That rather sounds like a bug to me.
I'd try browser automation to get the book pages, then OCR it (or just store them as images).