I guess there is a reason why it does things this way: Gnus can't know
  that the Gcc will fail before trying it, at which time the message has
  already been sent to via SMTP.  Maybe trying Gcc first would be an
  acceptable fix?  Does it have any downside?  I guess it will cause the
  reversed problem:  Gcc works but SMTP fails.  But that would probably
  be better, given enough warning so that the user knows the message was
  not sent.  Or?

To work, it will have to keep state about which has worked.  So
probably Gcc first, and delete the line on success, and then SMTP, so
then the nndraft: mechanism can handle queueing for both issues.

-- 
	Greg Troxel <gdt@work.lexort.com>