I think you want something that kind of "de interlaces" every 3rd line. Then fills in to make each one back to original height. The difficulty in this is that the scan might not align precisely.

I am wondering if it would be easier, to take three pictures of it at the angles required to see each frame sharply, then correct the skew.

Yeah, I thought about that, it can be done (making 3 photos I mean), the result might still be crap though, with weird light reflections, less sharp than a scan at best.

I am thinking that one could do it semi manually with a custom brush, that's like a venetian blind, and just block out the unwanted parts a bit at a time for each frame, then blend to fill in the whitespace the brush leaves when you stamp it. The advantage of that is that if the offset climbs a pixel across the image, a narrower brush, like 10% of frame size would allow manual alignment that's okay on each part of the image. IDK if I am explaining that very well.

I tried to zoom in my test scan a lot and the pattern is:
Line 1-2: first image
Line 3-4: second image
Line 5-6: third image
Line 7-8: first image
Line 9-10: second image
...

Just thought, there's gotta be some tools out there to merge two stereoscopic frames to make perfect single image... so could take 6 pictures of the source, frame one from two sides, up and down or right and left, then see what a stereo merge tool does. Then do the rest if it comes out well.