3 Commando Brigade

I can only speak to spaceborne applications and therefore reaction wheels (as most spaceborne rate gyros are ring laser gyros now), but the primary causes of drift were most certainly a result of resistance in the mechanisms, and thermal effects. If it was purely a case of being "wound up" by the earth's rotation, then a spaceborne gyro would drift at a far higher rate - approximately 240 degrees per hour in low orbit! Luckily, those effects are easily static-calibrated out if predictable, and can even be dynamic-calibrated out to an extent if you have other types of sensors to compare against.

Inertial nav systems are integrated for the exact reason that no sensor is perfect, either in stability of measurement (jitter/random noise) or precision (the less precise, the greater the natural drift from reality). By combining different systems with clever filters, you can build a mostly self-correcting system with a long stability time before it starts to diverge significantly from reality, at which point you can reset it to a known calibration point and continue

Of course the base drift is predictable. We know where the earth is and what it is doing. That is factored in to an INS and calculated when inputting the initial location data. I can't speak of laser ring and fibre optic gyros and their practical applications and limitations in space. Only brushed over them in trade training but the same principles stand, whether flying you to Ibiza, In geostationary orbit so you can watch babestation on your telly or hurtling off to a distant galaxy. 

Anyway. Thread drift.

Nick, I know you use trackir. Give this a go and compare the two. 
I know trackir is going to be far superior but I think you'll be surprised with the results. This obviously being much cheaper (free) than trackir.

I actually don't use TrackIR. I built a homebrew IR tracking solution with FaceTrackNoIR's IR tracker (hurrr), but wasn't ever satisfied with the stability of it due to its positional certainty and... the drift! xD

Sadly, the same reason I don't have TrackIR is the same reason my phone is a bit over 6 years old and lacking of the required OS or hardware fidelity to try this: money

What LEDs were you using? 

More than bright enough, I forget the exact serial. My issue with the FTnIR tracker didn't cope well with your head moving more than 10-15 degrees in most axes, and it didn't fail gracefully but instead would decide I'd flipped my head 180 degrees in that axis and was now spinning it further away from centre when I tried to return. It was an issue with the geometry measurement and prediction, and a particular sensitivity when it even came within spitting distance of an inflection point.

As I tended to be moving my head most during landing, I found that getting stuck staring over my shoulder at my right kidney was rather unhelpful and decided to return to alt-look scrubdom

Try filing the ends of the LED's flat as that removes the narrow focus from them. They have a built in 'lens' by way of the profile. This should allow them to emit at a wider field of view .

Start with sandpaper/file and move onto wet/dry and then emery to avoid too much scatter.

I'd give it a try, but I mostly disassembled the rig and it's now in pieces in my bedroom. I'm reasonably sure that the main issue was the software rather than hardware, so while your suggestion may give me a few extra degrees before the issues it'd still hit long before I wanted.

The principle application of reaction wheels and giros on space craft is to control their movement not measure it. You can see how that works here on earth. When a motorcyclist turns left, for eg, he doesn't actually turn the handlebars left.
What he does is put a force that would turn the handlebars right which causes the gyroscopic effect of the front wheel to lean left, thereby changing the direction of travel for the motorbike to the left.

This is great in space because it doesn't require rocket fuel to point the spacecraft in a particular direction. A Control Moment Gyro is generally accepted as the most efficient way to do this due to a higher torque output for a given energy input when compared to reaction wheels and torque rods. The big disadvantage of CMG of course is gimbal lock and would require a backup system to continue operation. 

Give me a shout on TS sometime