Good $3 MEMS gyros are about 100x better than that now - look at anything new made by Invensense in the past couple years. And their drift is pretty Gaussian-distributed, so the error scales as sqrt(n). If you combine 8+ of them on one board you can get about 5deg/hour stability...

Hm. Is it, though? If what you wanted to do was produce a large batch of calibrated IMUs, building a rig to do so wouldn't be an enormous undertaking.

Or do you mean to characterise the assembled vehicle?