How about trying your experiment with more recent books? The current crop are up in the 40000+ range.
OK, as a "sanity check" I went back and double-checked more recent submissions to see if they are no being formatted "reasonably correctly." IE, if this was an e-book that I had bought commercially would I say "yes, this formatted reasonably" or "no, this book has corrupted formatting." Not a high standard, just: does this book "work" or not? 40000 no 40001 no 40002 no 40003 no 40004 no 40005 no 40006 no 40007 no 40008 no 40009 no The two most common and glaring problems are: 1) Paragraphs are not formatted "reasonably" corresponding to any known standard of formatting at any point in time of mankind. And 2) No TOC in an e-book where one would reasonably expect and need a TOC.