Is there some way to rig up the stepper and the encoder by themselves, perhaps with something attached to the shaft that you can measure with a gauge?
So far as I can tell, a closed loop stepper is just a stepper, an encoder and some control software & circuitry. Which is basically what we have except that our control software is very basic. It looks like fancy versions also vary the current as needed and so they run cooler and quieter.
In any event, I'd be interested to see if absent the worm drive and the rest of the mechanicals whether that combination gives repeatable results on the 32000 test.