I think this is a really neat thing. However, I would need to use it in a different way.
I take lots of photos at home of records I collect, buy and (rarely) sell, and usually they come out slightly skewed, which makes cropping them a pain, inevitably at some corner something will get lost.
What I'd really like would be an interface where I could put the four corner markers on the image, in the place where the four corners of the photographed record are, and then have those transformed back into a square. So basically, it would be the opposite of what you're doing now. I know that it should be mathematically possible (even though my last encounter with this kind of stuff was a long time ago). UI-wise, it could be as simple as an "Inverse behaviour"-checkbox ...
With the current version I can do this as well, I currently draw a square on a new layer, and then modify the corner parameters of the underlying layer until the record fits into the square. But that's a bit cumbersome to do ...