Piggybacking client hello on STARTTLS

Not that I'm actually advocating this, but seems that we ought to at
least bring up the possibility of piggy-backing the TLS client hello on
the STARTTLS command in order to save a round trip.  Personally, I'm
against it because it would also involve merging the SMTP reply code to
STARTTLS with (perhaps) a TLS server hello in the response.  It just
makes something that was nice and clean messier without (IMHO) gaining
all that much.  Do others agree?

-- jeff