This looks good. A bit slow, but most of the pages are adequately recognized in automatic mode -- problems with some page numbers being clipped out, a few pages "recognizing" bits of the facing page, and a problem with blank pages that can be made moot by running Scan Tailor after merging the left and right views -- but it's sufficiently robust and sufficiently flexible that I'll probably stop using YAPP. I'll still be using ImageMagick -- Scan Tailor, as far as I can tell, only puts out TIFF files, and I still need to convert them to PDFs. Also, it's possible to bloat the original images by 10-20 times by choosing output parameters poorly -- color and 600 DPI takes a 1.5 MB JPEG and turns it into a 25 MB TIFF -- but the "mixed" mode does a good job of putting out crisp text and still preserving greyscale images.
I need to play with it some more, but I think this is going to become my main post-processing engine, at least until something better comes along. Thanks for the tip.