Well this may sound strange, but I think some of the best tracks (not mine!!) ever recorded turned out as well as they did because they were mixed from the very beginning - sometimes as early as while they were being written.
 
Please note, I didn't say ALL the best tracks, nor do I mean to suggest that waiting till tracking is completed will always result in a poor mix - some truly amazing tracks have been recorded with both approaches, but for me, having the final mix in mind has always worked better.
 
How can that be?

I walked into my first recording studio in 1974, it was an 8 track studio, and for the time it was a very well equipped 8 track studio, they had two Urei compressors (probably LA-3, maybe LA-4) and one Urei 1176 and two Orban Parasound equalizers. While I was working there they added a second 1176 and a pair of dBX 160s. They also had a spring reverb (I think it was also an Orban) and an EMT plate. The 8 track was a Teac 80-8, and they mixed to an Ampex AG-440, might have even been a 350, my memory is a little blurry. The mixer was, I think, a Mother's Finest Audio - can't remember the exact name, they were made in the Philadelphia area by former RCA engineers. It had a simple 2 or 3 band EQ on each channel, I think there were 16 inputs and 8 busses and a 2-Mix.

Anyway, that was where I started to learn the craft. Every studio that I worked in from then till the mid 1980s was, at best, an 8 track studio with similar gear. Eventually I stumbled into a 16 track studio and thought I'd found heaven, but alas all that really did was let us do more stuff in stereo.

What did I learn, and why do I still work this way (sorta)?

The first thing we would do was sit with the musicians and plan the recording. What instruments, how many vocal tracks, what, it any, effects - pretty much a road map. We'd take detours, of course, but we always started with a map.

So we'd know if we wanted the drums in mono or stereo, and we'd know there would be this or that so that the drums would need to be pre-mixed accordingly. I really can't recall a single session where we did not end up bouncing the drums down to two, or sometimes one tape track, so  we had to have an idea of what the final mix would sound like - we could make minor changes to the drum mix later, but we could not change the balance dramatically, nor could we remove a drum.

At which point we had six empty (?) tracks. We'd do the same thing for the rest of the rhythm section, we'd bounce guitars, bass, and keyboards down to a stereo pair, so all panning decisions and basic balance had to be made right there. Sometimes, for musical or scheduling reasons we'd record the entire rhythm section together (I always preferred that). So we'd have to really pre-mix the drums, like in the mixer. That was a challenge, but the performance always seemed ever so slightly better, so we did it.

Now we had four tracks left for vocals, and we'd end up bouncing down the background vocals a couple times. At which point we would have:
one stereo pair for drums
one stereo pair for the rest of the rhythm section
one stereo pair for the background vox
one track for the lead vocal
one track for whatever else we wanted
 
If we had horns we'd end up bouncing the entire rhythm section down to a stereo pair to free up a stereo pair for the horns. And if we were doing anything remotely different we might have to work with mono tracks for the drums or even the entire rhythm section.

Needless to say there was some stress involved.
 
When we moved to 16 tracks things became somewhat easier. A typical (as I recall anyway) session might be mapped out something like this:
1) Bass
2) drum overheads L
3) snare/hat
4) drum overheads R
5) lead guitar
6) acoustic guitar
7) keyboard mix  L
8) keyboard mix  R
9) hand percussion
10) lead vocal
11) rhythm guitar
12) background vox L
13) background vox R
14) piano L
15) piano R
16) Bass drum
 
You can see that we had the ability to push off some decisions till much further down the road, and that was kind of cool, but we still had this discipline of thinking ahead firmly ingrained, so we tended to think that way even with the extra tracks. And there were lots of different version of that - sometimes the drum overhead was mono, sometimes all the keys were pre-mixed, sometimes the piano was  mono,  sometimes the rhythm guitar was stereo, and so on. But it was a big leap.
 
I still think that way today, even though I have, for all intents and purposes, infinite tracks and infinite processors and effects. The effort to plan things out keeps my mind on the right path, I think.

I've made some minor tweaks so that  I can make major changes at mixdown time - I'm not a idiot<G>!
 
I still print all my processing and effects while tracking, but I keep the unprocessed tracks too. So if I change my mind I can swap out one phase shifter for another, or re-equalize or whatever I need to do.

It's kind of the best of both worlds! And I apply all my processing and effects to live and MIDI or VI tracks.

How can this help?

For one, I'm always thinking about the mix, how different instruments will fit in to the final mix. I find that really helpful.

For another, at mix time the only thing I need to do (in theory anyway) is ride levels and add the final reverb. Believe it or not, sometimes that really happens.

Give it a try, I think you'll find that with a little practice it not only makes mixing easier, but it also makes the structure of the song stronger.
 

Recording with effects depends a little on the style of music you are producing.  For example working with samples as Bats mentions and recording with effects may make it difficult later on.   Standard rock music style here.   Adding effects later on allows you a lot of control.
 
But by the same token say you are producing electronic music for example things are different.  I have 8 or 9 hardware synths for example and used to in many cases turn the effects off before tracking.  But then you end up with similar sounding effects being applied later on to the same machines which is good but after a while I decided to try tracking with the synths effects being applied. 
 
What happens in this situation is you end up with the parts sounding different to each other and in a funny kind of way they stand out more so.   All my synths have reverbs in but they all sound a little different to each other.  The built in reverb in a Kurzweil K2000 sounds different to a Roland JD800 or a Roland JV2080.  Often with these patches the effects play a very integral part of the patch so switching effects off on them can actually reduce how effective that patch sounds.  By leaving them on those instruments tend to sound more different to each other and in a way they stand out more from each other later in the final mix.
 
The trick I find is to simply edit the effect slightly and usually by that I mean backing the reverb off maybe slightly to the point where it is still there but not so much now e.g. less swimming in reverb. You can get quite good at judging this and even if you backed off a reverb a little too much, it is easy to add a touch again later with a plugin say.  It will usually then give you the amount of reverb you want to hear in the end and at the same time add interest to that patch. 
 
I used to own a Yamaha SY77 synth.  Many of the patches were swimming in too much chorus and reverb.  Often switching off the FX made the patch more interesting and detailed.  What I ended doing with that machine was often backing the reverb off down to that just being heard type of thing and either turn the chorus off or alter the wet/dry mix in order to get a much nicer balance.  That is all it took and then the patch still sounded lovely but had the effects on at the same time.  Often later in the mix it rarely needed any more processing. 

Reading Bill's post reminded me of how I used to work in the early 80's.  I only had a 4 track machine (with Dolby C noise reduction) and a quality two track machine.  Firstly I would limit the end mix to only contain 9 parts.
 
I used to track the first 4 parts on the 4 track.  (punching in and out)  Then mix that down to stereo and transfer to the two track.  Adding in effects and also adding in a 5 th part live.  So the 2 track had the first stereo stem on it.  Then I would wind the 4 track onto new virgin tape.  Transfer that stereo stem (with 5 parts on it) to tracks 1 and 2 of the 4 track while add in live a 6 th part.. 
 
Then add in parts 7 and 8 on the 4 track. (punching in and out) Then mix all that down to the stereo machine adding in the 9 part live.
 
I ended up with a full stereo mix that contained 9 parts.  Those mixes to this day sound seriously good and perfectly balanced etc.. And noiseless as well.  They do not sound like a 4 track machine.  They sound big and wide and quite spectacular in fact) I was forced to do it this way then before I moved to 8 track.  It proved that you can actually do it.  Limiting yourself to get all your ideas down to 9 parts is also a challenge.   If I really needed it I could do it one more time back and forth and get 13 parts which I did for the much bigger tracks. 
 

Sometimes I think the creative effect chain is essential to your creativity and playing. Like I don't know many people who would record an electric guitar DI clean without at least monitoring the amp/distortion. So if you wouldn't do that for any other instrument or synth either. And so if reverb or aggressive compression or distortion is essential to the mojo of what you are playing, keep it there. If you need to hear your side chaining effect or delays to record your next part, keep them. As others have mentioned, make sure you can make changes (fx can be bypassed, DI was recorded too, etc).

To reconcile some of the other feedback above, I split Fx into two categories in my mind: essential contributions to the soundscape, and mix "adjustments". I don't bother with the latter really, as any balancing eq or compression I do too early usually isn't contextually appropriate any more when i am finished tracking and i have to redo it once I have the full track count in front of me. On the other hand, prominent delay effects, or saturation, distortion, etc, to get my desired tone are so essential to the artistic direction of the song that I absolutely keep those on, and meddle with them endlessly while recording other parts, as I drive towards my vision for the track.

cboshuizen
Sometimes I think the creative effect chain is essential to your creativity and playing. Like I don't know many people who would record an electric guitar DI clean without at least monitoring the amp/distortion. So if you wouldn't do that for any other instrument or synth either. And so if reverb or aggressive compression or distortion is essential to the mojo of what you are playing, keep it there. If you need to hear your side chaining effect or delays to record your next part, keep them. As others have mentioned, make sure you can make changes (fx can be bypassed, DI was recorded too, etc).

To reconcile some of the other feedback above, I split Fx into two categories in my mind: essential contributions to the soundscape, and mix "adjustments". I don't bother with the latter really, as any balancing eq or compression I do too early usually isn't contextually appropriate any more when i am finished tracking and i have to redo it once I have the full track count in front of me. On the other hand, prominent delay effects, or saturation, distortion, etc, to get my desired tone are so essential to the artistic direction of the song that I absolutely keep those on, and meddle with them endlessly while recording other parts, as I drive towards my vision for the track.

I agree with this, although sometimes I find doing the mix adjustments as I go along can serve as a sanity check to what I've written.
 
Sometimes I get 2 or 3 parts that musically work together, but in the mix they just hog each others frequency space. I might try EQ to thin things out, different guitar/synth sounds, or play pads in a different inversion or octave, but if the parts are causing me mixing issues early on it's time to rewrite some parts.
 

Why not provide both the dry and affected tracks to the mixing engineer?
 
If I love a guitar sound, i'll make a dry backup and then hard-code the sound I want.  USUALLY my mixing engineer has used the sound I wanted, but once or twice it hasn't sounded "right" in his/our mix.  Then I'll give him the dry track and we can affect it to match the full mix.

The same applies to synth sounds.  Sometimes the effects are just so bedded into the overall sound it would be silly to remove the effects.  They can be obvious and overdone, but they can also be subtle and baked in such a way that they are hard to hear.  Until you switch off the effects then the whole sound can change.  If the total sound is inspiring you in a way then leave them in.
 
Some synths have multiple outputs too and you can route dry sounds to say the main stereo pair and the effects to an auxiliary stereo pair hence allowing you to track both things at once.  Giving you options after on. 
 
I find with recording guitarists a good thing to do is to record the sound coming directly out the guitar. e.g. using a DI.  Plus the sound at the end of the effects line e.g. a second DI and the total sound coming from the speaker.  Then you have all three options.  The speaker sound is often used as they say but having the original guitar sound can also be a saviour at times especially with all the amazing effects and amp/speaker sims that are out there now.  Some guitarist can overdo the reverbs and distortion and it is nice to be able to rebuild this sound in a more controlled way.
 
 

If you are using effects as part of sound design (i.e. reverb or slapback to get a certain guitar tone, or any type of effect for a synth) they should be printed to the track.  But record it into the track and commit to it.  Most of the time if I see FX tracks delivered as part of the stems, I discard them, unless there is a good reason to keep them.

the only thing you really need to remember about tracking......
 
THAT is where the magic happens.
 
not during mixing.

yeah, effects can affect your performance
for instance, reverbs while laying down vocal trax can change phrasing
read that lennon was obsessed with that