First of all, it's an easy experiment to unbalance a mount, so give it a try and see if you get consistently better results by biasing the mount. Don't forget that you will have to stop (if it's a GEM) and rebalance it when you cross the meridian every night.
As I said, I find it probative that there is not a peep about doing this in the PHD manual or in any manual that I can find for setting up a GEM. Here's an excerpt from the MaximDL manual "Backlash on the RA axis can become a problem, though, if the mount is poorly balanced, e.g. heavier on the west side". To me this means "balance your mount" for best guiding.
The worm is turning the mount using the gear it is meshed with. To rotate any object is to change its direction. To change the direction of an object requires that a force be applied to it. So, in a balanced mount you are never going to be "braking" with the worm at all. Not only do you need to apply force to rotate the system, but you also need to overcome friction. So, I just can't see that the worm in a balanced mount is going to be braking on one side and "lifting" on the other. That implies an imbalance.
So, I think my position is that if you find that you need to unbalance to get better guiding then there's some not so obvious reason for it, but it's not backlash. So, my guess is that what's actually going on relates to the difficulty of accurately balancing a mount under all possible load configurations, particularly when there is considerable friction present during the balancing process.