Hmm interesting analogy. I agree that you should do it physically where possible but damping high f noise from a glow motor running at several thousand revs is gunna be especially difficult (to do mechanically). On the other hand there are many other approaches (maths, software, etc) which;

1) Consider that high f noise (above some threshold say) could not possible be coming from the gyro (as these only updata at like couple hundred hertz maybe (im just guessing))
2) Consider low f drift and adjust accordingly

Kalman is not the only way, and its probbably not the absolute bestest way but it wrks and is (relatively) easy to understang. Go to rotomotion.com where they sell IMUs and give avay Kalman C code. As for assembly- hows ur floating point assembly skills ;-) ???

Also the main thing is the more you try and do with these Df's (in terms of autonomous flight control, hovering, landing, UAV etc) the more important the sensors. Absolutely and totally spend 90% of your time eveluating the best sensor package that gives you the data that you need. I believe that means NOT re-inventing the wheel- universitys, hobbyists, companies, have done all this stuff before; use their advice in terms of sensor choice (IR vs sonar vs gvro/accel etc). There is infinate flexibility in what you can accomplish in software with the right data. This means (if ur interested read this)

Consider the maths and physics, its essential because then youl realise that to control certain parameters; position (used for landing etc, path planning), acceletration (hover, landing), velocity (hover, landing, path planning), you must have the sensors in place to sense this.
Anyways thats the last im gunna say on the theory of it really (unless someone asks me).

good luck