My project has 29 instruments, and is often fairly dense. It's also over 20 minutes long (so far). This week I've decided to step back and work on improving the mix so that I can hear it closer to the way I will ultimately want it, so I spent all yesterday EQing every track, changing the synths a bit, adding buses, etc. So far so good.
 
This project uses a chamber orchestra, and has a pretty wide dynamic range throughout, and the subset of instruments playing at any given time varies tremendously. I might have the strings and harmonica for two minutes, then it'll suddenly be brass and timpani for two minutes, etc.  I realize that I need to spend a bunch of time now fiddling with levels. Given the fact that this isn't like a pop song where one can mostly just set each track's level and call it good, I suspect this is going to take a huge amount of riding levels of every track at any given time. 
 
So, I'm wondering what's the best approach for accomplishing this without painting myself into a corner or otherwise doing something stupid. Keep in mind that this piece is only half done, and I'm taking a little break to do some mixing before I go back to composing. So a solution like "freeze every track and export it" isn't practical. They are all instrument tracks, so I can put a volume envelope on each one. I also have each track output to a bus for their section (strings, winds, etc.) so I could put envelopes on those section buses too. 
 
Is it really as simple as "add a volume envelope for every track and start editing nodes"? Or perhaps it's better to set levels for each movement separately, or do a single instrument beginning to end first? Should I start with the piano, or the drums, or the strings, or what? I know there's no one answer, just looking for advice. Thanks!
 
 

Great mixes use a truck load of automation.
Brew yourself much of your favorite liquid stimulant, lock the door and get after it.

So you're telling me that even if I were doing this day in day out for years I'd still approach each project in kind of a come what may fashion?

I have drawn in volume envelopes for years, but lately I will use a Mackie MCU-Pro and ride the faders in "touch automation". It is often helpful to put critical tracks next to each other, so they will be on consecutive faders. Also, moving the active faders on the MCU to the subgroups may be a good thing for orchestral recording, if you subgroup the sections. If you mess up or want to update a section, just back up, or go to the section and re-write the touch automation. I like this way better lately, because I can close my eyes and intuitively mix the tracks vs. drawing the envelopes in in a more clinical, microscopic way.

I do most automation like this

 
After doing hours of automation you get better at it.
You need to establish a base mix first (all levels approx. where you want them) then move on to the more detailed work.

I don't know about "best" practice, only what I do...
 
Assuming your composition is MIDI (i.e. you don't actually have a 60-piece orchestra in your basement), you can use both MIDI and audio automation. What I like to do is stick to MIDI automation during the composition phase, and to paint changes in broad strokes, e.g. "tympani up in this section, push the cellos back in this part". I find it easier to do the fine adjustments (e.g. quick swells, mutes and accents) on frozen tracks, where I can better see what's going on. And it just makes more sense to do all your final tweaks after the composition is complete anyway.
 
What I almost never do is use a hardware controller or onscreen knob/fader to record automation. It's nearly always hand-planted: double-click to create a node, right-click to choose the curve. That avoids cluttering up your screen with thousands of unnecessary nodes, allows much greater precision, and is infinitely easier to edit later. 
 
I also use bus automation a lot. Once the individual sections are balanced and weaving in and out nicely, subsequently combining them into busses lets me automate them as macro-sections, e.g. separate busses for strings, brass, woodwind and percussion. It's a time-saver.

You should definitely get into the practice of using buses to group like instruments together. Mixing a large project with more than 50 tracks is just not possible to do and maintain some objectivity. You'll constantly be pulled into the minutia of scrolling around looking for tracks that need attention.
 
Start with a good solid drum and bass mix, and build the rest of your mix around that. With practice, you'll begin to realize that you don't need to treat every track individually. For example, unless you are redoing "Bohemian Rhapsody", you can probable mix all of the background vocals of a track together by simply adjusting levels and basic EQ corrections as needed. Bounce them all down to one stereo track and then do the rest of your processing on that in the context of the mix. Same with guitars, synth pads, etc. Similar instruments will generally need similar processing, so bounce them together after leveling them and fixing issues that are track-specific.
 
Once that is all done, you'll have a much smaller mix to deal with.
 
I will say this, I don't bother trying to start mixing until I'm done tracking. It tends to get in the way of the muse...
 
Regards,
Dan