After recording I normalize all tracks so that all levels are in the same ballpark.
But I am always torn as to what max level to use. 0.0dB is tempting but bad for so many reasons.
So what should one aim for? I believe Redbook states max levels for CD but
that's really only a consideration when mastering. Do others face this dilemma?
Any recommendations?

My recommendation on this would be to suggest that you might be dealing with the "problem" in the wrong place!?  I would try to focus on getting your gain-staging set properly going in to your recorded SONAR track rather than to process (normalize) all of your tracks.  I'm sure that others may chime in on this as well....... but that's my "two-cents"!

Cuitlahac's advice is sound. However if you are going to normalize, leave some headroom - at least -1 dB. For acidized loops, I'd recommend at least -3 dB as the stretching process can produce additive phase relationships.

His method is one I have used for a long time now. I sometimes  tool copy each track into Wave Lab to accomplish this as it has better tools to fine tune the process. Like you can also deal quickly with random peaks manually which can eliminate using a peak limiter. I use Sonar's Normalizer often on tracks that are tame but just a little too low in level. All you nay sayers that say this is destructive? Phooey. 
 
Proper gain staging is fine when you are working alone and have time on your hands. This is best practice! I certainly roll with that. 
 
But in the case of live band recording or when your in a session with clients who need to work quickly,  it is wise to record at "safe" levels. Digital audio doesn't care as much about hot signals.  Normalizing to -2 or ? will  solve this issue and give you a track that is much easier to work with. Sure you can just turn up the Gain of the track, that works too. But the option of normalizing has worked best for me. 
 
So to answer your question, do not use 0,There is a danger of overs at 0, you can use anything less than that. 
My little trick is to normalize tracks to a target of where they will sit in the mix. 
So I might put lead vocals @ -2 and backing vocals to -6. 
My goal is all faders at unity or close. 
This is the way I do live sound too. 
I use the input gain and EQ to set most levels so that my faders are at unity.. just me and how I like to work. When I see a fader down at the bottom or jammed at the top I know something is wrong with that channels gain stage.  
Mixing is understanding this completly. Input / track levels also interact with all your Aux/ effect send levels so a weak track will not drive a reverd buss the same as a hot one as example. 
 
 

Cactus Music
My little trick is to normalize tracks to a target of where they will sit in the mix. 
So I might put bass at -4 and backing vocals to -6. 
My goal is all faders at unity or close. 
This is the way I do live sound too. 

More good advice. Hopefully sometime in the future I'll play someplace where you're doing live sound

-0,2 dB.
We digitize for thousands of clients tapes and vinyl to CD and hard drives.
The best level is - 0,2 dB. Some CD-Players don't like 0 dB.
You also have no problems with CD-Masters by 0,2 dB.
But in most case we do not normalize - instead of this we take a limiter (in our case the Waves L3 UltraMaximizer) as the last plugin (in a very small range of less than 0,5 dB). To ensure that the audio material never reaches 0 dB.
 

Sounds like I work much like Cactus. Time pressure, live band recording, no idea what's going to happen on stage, no second chances. So generous headroom is priority #1.
Gain-staging is out because I need to exchange tracks with collaborating partners. And it's time-consuming.
 
In the old days I normalized externally in Sound Forge/Wavelab/Audacity before I discovered Sonar has normalization.
 
As Cactus also points out, peaks are a problem here, too. I find expecially in a small non-distortion based band (typically 3-piece jazz/blues outfits) waveforms may momentarily reinforce to produce a fluke peaking half-wave that messes with normalization.
 
Thanks for the tips. So no hard-and-fast rules, but 0dB is clearly bad, and, I shall revisit Wavelab.
 
 

THambrecht
-0,2 dB.
We digitize for thousands of clients tapes and vinyl to CD and hard drives.
The best level is - 0,2 dB. Some CD-Players don't like 0 dB.
You also have no problems with CD-Masters by 0,2 dB.
But in most case we do not normalize - instead of this we take a limiter (in our case the Waves L3 UltraMaximizer) as the last plugin (in a very small range of less than 0,5 dB). To ensure that the audio material never reaches 0 dB.

That would be for finalization/mastering, then. Thank you, your advice greatly appreciated.

-.02 is for mastering and I don't think Sonars Normalizer can be tuned that fine. For tracking  -1 I think is the closest you get to zero. Not at my DAW so cannot conferm.  
 
 
 
 

http://forum.cakewalk.com/Adding-programs-to-the-SONAR-Utilities-Menu-a-new-tool-Updated-to-v23-m3237117.aspx
 
If you use the new app it's easy to add a wave editor to the tool menu. 
Then you just highlight the track and it will open in the wave editor for fine tuning. 
I'm still hoping Cakewalk will update the wave editing options. 
The two tools I'm missing are analyzing tools, Global RMS level and Peak level. 
In Wave lab I can look at what a tracks peak level is before normalizing. 
And RMS average is also super important for making calls. 
These two tools are my most used of all. 
If a track shows it's at 0 db peak but -22db average RMS then you immediately know you have a few peaks to deal with. 
 
Thanks Mr Anderton, It would be an honor to do sound for you. and if I do, I'll make sure your the loudest...  :)