Your questions intrigued me so I had a look at Xsane once again. I
first tried it casually a couple of years ago but didn't find it too
useful. This time was worse! Sane uses gocr found under the file menu
on the scanned image viewer. You have to install gocr separately. The
accuracy was only about 50% - but it should be close to 99%, which
still means many corrections per page.

I have no doubt that with a lot of fiddling and perhaps a different
ocr engine the accuracy would be improved dramatically. I just offer
you this as a "proof of concept" that it does work.