In our situation (2:1 ratio belt driven preloaded ball screw, gibb based Bridgeport Eztrak mill with with high $$ true ball screw support bearings) we found the "error" contributed by the OEM "gilmer" style square tooted timing belt to be inconsequential.
Granted, we spent a small fortune on the ball screw bearings and weeks tuning and adjusting the machine to remove/eliminate slop. During the tuning process, we also spent a ton of time playing with belt tension and the like.
A belt in good condition had virtually NO effect on net accuracy once we got the slop out of the rest of the system. The BIGGESt contribution came when we had a pro tune the servos so as to get the follower error of the servo's "tuned out".
We tested via cutting perfectly round circles via a canned circle milling program. We then checked the form for roundness on our camshaft measuring machine (measures to 0.0000025" resolution).
When the machine was properly tuned, the circles were round within 0.0003" or better and the "flats" at the 4 direction changes were 0.0001" or better - the "flats' are nearly IMPOSSIBLE to eliminate due to gibb slop needed to allow the table to move without binding. We ultimately learned that the above "errors" were due to other hysterisis issues that simply weren't practical to resolve on a gibbs based "bridgeport mill".
Does the average guy need such accuracy? No.
But we CONCLUSIVELY showed that "belt induced errors" are much less in reality than what one might "expect" IF you have a well adjusted/tuned system made of parts with decent integrity.
Worn/tired crap isn't worth spending the time on. Get good sprockets and premium belts, properly tension them and make sure they "wrap properly" and spend your time and money on other stuff that provides a better cost/benefit ratio - like a good servo/stepper tuning regime and the best damn ball screws ans ball screw bearings that you can hardly afford.