You say regular intervals... is this frequent enough to have confidence that a test should/should not present the problem in a predictable manner?

If your 9060 supports pluggable connectors to the steppers, I'd swap X and Y, and - with due care - run some g-code to present a test engraving that will demonstrate the problem (e.g. repeated iterations of a fixed size square) - any lost steps will present themselves clearly there, and design the g-code to avoid any limits.

If the problem remains on the Y-axis, then I'd follow cropwell's advice.

My thoughts are it's likely to swap with the connector swap - indicating a problem within the controller/computer setup.

If you've got access to a 'scope I'd be examining the signalling going into the stepper driver, just to characterise it... but that's just paranoia on my side due to experience with a previous motion controller.

Edit: Oooer, just noticed the picture (!) - that showing what looks to be a similar number of steps offset on each successive pass in Z?, is there evidence within the rest of the workpiece where the problem is introduced? Is this a problem with a new design, or can you repeat the problem with g-code that previously worked?