[bojan]

OK, I have done my homework over weekend, about what & why people are modifying on their eq6's, and what is the outcome, in numbers

Summary of reasons for mod:
1) Sometimes large Periodic error
2) Large backlash (not good for autoguiding)
3) generally problematic quality of assembly at the factory.

The periodic error comes from couple of sources:
- Main worm gear (30 arcsec p-p)
- other gears in the reduction chain (1-5 arcsec p-p) and bearings (dirt in them, lack of lubricant etc)

So, belt drive is done mainly to have faster slewing when GoTo, but since the reduction ratio is significantly lowered, microstepping must be used (Bartel's software handles this nicely). The belt drive itself will not improve the PE significantly, and this is to be expected, because the errors introduced by gears closer to motors are reduced by the subsequent reduction stage anyway, so the only gear without further reduction is the main worm gear and therefore it remains as the most significant contributor to PE.
Also, belt drive reduces backlash, as all other contributors (apart from main worm gear) are removed.

To me, the biggest problem is inconsistency of test results for PE... it seems that either quality of mount varies drastically from specimen to specimen, or that the measurement results obtained by their owners are questionable. Obviously, I will have to perform my own tests and see how it goes, and if modification is necessary or not.

Outcomes of these modifications and refinements are also not consistent. Some people achieved <5 arcsec p-p with the same main worm gear, by only dismantling, cleaning and putting it back together (with more or less adjustments done to the relative position of the worm gear and worm).
Some others even changed the worm gear assembly (there is a company in Germany [Gierlinger], supplying much better quality replacement), but without significant improvement (?).

To my understanding, the error should not significantly exceed 1 pixel p-p (that assumes ideal optical system.. star images are always much lager than this).

So, for my 10" newtonian and 400D in prime focus the pixel size expressed in angle units is ~0.8arcsec. Very narrow (it will be a bit bigger with coma corrector)... Something like this is impossible to achieve without good autoguiding.