Organizing a layered animation presents the same challenges. If there are too many layers and too many detailed tweaks, the scene becomes incomprehensible -- but too few, and all the cons of working with dense baked data start to hold you back.
You can override your baked data with IKs and constraints, but if you start to think about layering the animation on those as well, you'll run the risk of making your head explode. When do you stop?
There's no single right answer, but there are a couple of good questions to consider when organizing your layers.
1. Will you ever touch this again? If you've made a tweak and you're withholding artistic judgment to see how it plays out, you probably want to keep that layer around. If you've just fixed a technical glitch, though, you might want to merge your fix down to clean up your workspace.
2. Are you planning on retiming? One limitation of working with baked data that layering doesn't really eliminate is the fact that baked keys don't stretch as well as sparse keys with nicely tuned tangents. If you think you'll need to lengthen your animation, you'll probably want to keep your layers around longer to make sure you don't throw away higher quality data.
3. Are you sequencing or sketching? If you're using your non-linear animation the way it was originally designed, you'll be stringing together multiple clips with blends and sometimes overlays. You'll want to keep your clips separate until you're really happy with the gross timings and blocking of your scene.
On the other hand, if you're using layers to rework a single sequence -- for example if you're trying to lend some life to a lackluster captured cycle -- you'll probably want to add and delete layers frequently as you sketch out the effects you want.
In this case, baking frequently has the positive side effect of helping you spot hitches, since it forces all the overlapping clips into a single set of f-curves, where it's easy to iron out small irregularities by deleting a few offending keys from one place.
Perhaps after all this, you're still unconvinced. The animator's horror of over-keyed curves is a hard thing to shake. If that's the case, you ought to consider using some of the tools we've talked about to do blockings and establish key poses for conventional animations.
The talk that Jeremy Yates and Judd Simantov of Naughty Dog presented at GDC 2008 gave a detailed and very illuminating look at how the animators of Uncharted: Drake's Fortune used mocap data as a foundation for more traditional animation.
They built tools for grabbing poses and rough timings from the motion capture data but heavily tweaked the results so that the overall flavor was more like hand animation than straight captures.
Without committing completely to an all-layered approach, they were able to simplify and bulletproof their rigs by treating IK and constraints as add-ons rather than permanent features of a monolithic rig. They got a lot of the virtues of a layered approach -- flexibility, creative freedom, and speed -- without giving up on the virtues of careful hand keying.
Even though your animation needs are almost certainly quite different, their experience illustrates how much power can be had when you give up on the quest for perfect rig and settle instead for a pretty good one combined with pretty good layering tools.
We must admit that reports of the death of rigging are a trifle exaggerated. Rigging might not be dead, but it's not quite the only game in town anymore -- and anything that busts up the logjam in animation has got to be a good thing.
So take a break and try to teach yourself the Maya Trax Editor or Max's Animation Layers next weekend. You'll probably be a bit baffled to start with, but once you see what new tricks you can do, you'll be glad you spent the time.