By "do the job well" do you mean ease of use and effort or what individuals produce due to lack of competence!
By "do the job well" I mean having features that are well-matched to what book coders actually need to code, and when book coders want to produce something that renders "correctly" on an HMTL, EPub, and/or Kindle device that rendering well-represents the efforts of the coder, and can do so without the coder having to resort to extraordinary tricks, hacks, and work-arounds to accomplish common book coding tasks. There a quite a number of books out there that intelligently regale the problems with trying to use HTML to code books, as opposed to HTML sites. For EPub see the books by Castro, Deuchler. For Kindle see the book by Tallent.
Like I mentioned you have to encode as close to the original without interpretation, and later change or ignore it during further processing.
Somewhat agreed, but the book coder always has to make *some* decisions to avoid simply slavishly creating a PDF/bitmap photocopy -- which is what Google Books is *already* doing, for better or for worse.