I think that is a very salient observation. To reach its full potential the unit needs both pixel occlusion (which is knocking out the light your eye would normally see with light you generation) and pixel registration (which is keeping the pixel in the same place in 3-space outside of you). All of the demos focus on occlusion (the creatures seem like they really are in your room) and skate over registration accuracy. If you were going to tall someone how to fix something, registration accuracy would be critical (for example)

Also people are very "forgiving" if things seem to wiggle a bit in the world but looking at the Google Tango device this weekend I got to see some of the challenges of keeping registration of the pixels high. Its frustrating if your ruler end point is "jiggling" a bit close to the spot where you want to start a measurement.

So without good registration you get eye fatigue like you used to get with pilots flying long range on bumpy flights constantly looking at instruments. Only in the case of pilots they can look outside to the clouds (which don't "bump" in space).