On Wed, 14 Apr 2010, Gardner Buchanan wrote:
While I agree more with BB's conjecture, than Don's I have seen
no real statistical evidence on either side.
My own experience, which is very old now, is of encountering
titles in Palm-compatible formats that had manifestly been derived
mechanically from the PG plain-text versions. This is just
an anecdotal point, but it matches BB's "eucalyptus" data
point.
Another couple of anecdotal points. There are two paper publishers I've worked with a bit. Not recently, but a couple of years ago. Both had scripts to take the plain text and allow them to typeset in a couple of hours. They didn't use the html ever because it threw too many exceptions that required hand input to resolve, and therefore took a lot longer to get typeset. The proofread after to make sure nothing got messed up took longer.