Kovid, I'm happy to open a ticket but I can't really attach the PDF as it's a commercial one with the DRM removed (by someone who knew what they were doing). It's also large - about 7Mb.
Could you do your tests with a small selection from one or more of the debug stages and/or one of the epub sections? There were about 10 of these - only 3 with the problem tags.
If so perhaps you could specify a combo which would suffice and I'll do my best.
In the meantime I'll see if I can find a way to extract one of the PDF's problem pages without destroying the problem. Any suggestions for free software which will do this gratefully accepted.