I was going to just ask the question in the title, but I decided not to be a lazy beggar and had a go - even though I could just be "rediscovering" something that everyone and his uncle have been fully aware of for centuries. That happens a lot to me...

I used a manky old butterfly foot that was already on the rig and shot with a 50x Mitty on a 135mm tube lens for 33x onto the sensor. FF Sony A7r2 running in APS-C crop mode (18 megapixels) with around 200 images per stack (nominally 1 micron steps).
First, a straightforward stack. Flat lighting, no retouching, and the stack didn't go all the way through. Enough "sticky-out" details to detect a stereo effect if it works though

I figured it would be worth creating a baseline stereo to compare against (and if I didn't, one of you would ask for it) and created a stereo from the image above using Zerene. I used +4 and -4 offsets. Worked as expected.

Then I rotated the subject 4 degrees and took another stack. Well, I meant to rotate it 4 degrees, but as well as rotating, it lurched left and seemed to tip sideways and forwards a bit too. My rotation stage is very scratchy and coarse - not really designed for such fine control. Anyway, I left it as it was and stacked it again.

So now it was time to see if I had a working stereo pair, but I was sure it was doomed as the "lurch" would have messed it up. Did it work? You be the judge...

One big benefit, despite appearances, is that this method can use fewer images in total (less than half as many in each stack) to generate good stereo results for screen resolution output. Zerene's stereo algorithm starts to get "artifact rich" if the stack is too sparse. Encouraging...