Using EKOS's built-in auto guider, is there an advantage to using a specific number of calibration steps?

I'd suggest that using more steps probably results in a more reliable calibration. I rarely recalibrate, but when I do, I use 10 steps. I believe the system will terminate the calibration phase anyway when it moves about 25 pixels or so.

I guess the main downside (besides time) is the potential of driving the guide star off the image, if it happens to pick one near the edge (which is discouraged).

In my workflow, calibration is done supervised (with me watching), and I don't mind if it takes a few minutes to get a good calibration. Then I re-use the calibration (so, no supervision) until something happens that causes it to be out-of-date (e.g. rotating the guide camera).