Well that could be explained by backlash between the worm and wheel, so if you turn in one direction and take a cut, then say went a bit further round, but then reversed to take another cut, you wouldn't be at the same angular position. My rotary axis is based on a Myford dividing head, which has adjustable friction so the axis isn't deflected by cutting forces; and I use cutting strategies that make sure that all angular moves that count towards actual cuts are made in the same direction. You can turn in the other direction, say to get back to a starting position, but if so need to overshoot slightly then go in the "right" direction to actual zero. Your controller probably also has backlash compensation but I'm not sure how effective that would be on a rotary axis.