Hi Bsmith,
If you are really concerned about ultra-precision over long distances, I would recommend a system that uses linear scales for feedback. This sidesteps all the concerns that you raised above.
As far as I can tell, there is no guarantee of absolute position with belt drives of any sort. They tend to be quite repeatable, I would say, if the load is constant, because this means the stretch factor will always be the same. The accuracy of a common timing pulley drive is dependant on the precision of the outside diameter (actually called the pitch circle) of the toothed wheels. Contrary to appearance, the toothed belt does not provide the accuracy, it merely keeps the belt from creeping on the surface of the pulley. Whatever degree of freedom that exists because the belt tooth is narrower than the space cut in the wheel is the potential backlash amount. Belt tension tends to confuse this issue of timing belt backlash, because there may be sufficient tension for the belt to drive from the pitch circle, instead of "pulling hard" against the teeth of the wheel.
One advantage of the belt system, is that more often, better means are used to fix the individual driving members (the pulleys) to the shafts. I am thinking of taperlock bushings in particular as being the ideal fastening method.
Compare this to a direct drive, which in the cheapest instance, uses a straight solid coupling: but the keys and keyways are a slide fit! Potential for hidden system backlash is quite high, should the setscrews ever work loose, and they often will, on a system that is working hard.
So if you use a two piece direct coupling, then you have to be concerned about the compressability of the joint used in the coupling.