Peter,
the problem with encoders on the motor shaft is because the only problem you are solving by this approach is eventual motor slippage (or missing steps if steppers are used) and you have the correct tracking rate and position at this point (motor shaft) in case of DC motors (closed loop takes care of this).
But, the rest of the gear reduction (or timing belt) and worm gear errors are not compensated for periodic errors and backlash. So, the closed loop with encoders on the motors will be OK for GoTo - but still not adequate for accurate tracking needed if long focus optical systems are used for imaging (for example, I found that +-30arcsec PE error is acceptable for 300mm lens, up to 30 sec exposure time - precise enough for what I do (photometry), so I don't use guiding yet).
Sure, there are precision-made worm gears (Byers for example, see here:
http://www.cloudynights.com/item.php?item_id=2299) , but they are very expensive, and still not completely free from PE.
Also, you should not forget about atmospheric refraction, and not ideal mount polar alignment and flexures (people often tend to overload their mounts).
All the above can be adequately compensated for by guiding, for the fraction of the price to be paid for any other approach to this problem. Even the lack of suitable guiding stars in the area area can be offset by accepting somewhat larger field rotation as result of using stars away from target (better mount alignment, the smaller the field rotation will be).