Using gnus from CVS about a month ago I forwarded one of my own signed
messages that itself contained a forwarded message, and a colleague
pointed out that the MIME boundary string had been reused.  That
particular example can't be sent to this list, but I managed to
reproduce the problem with test messages.

The inner test message had boundary =-=-= originally.  That got remapped
whent he message was forwarded, which I suppose is ok.  But it got
remaped to ===-=-= and that's the same value that was chosen for the
outer message's multipart/signed.

The below was generated with current CVS head epg and gnus git as of:

  commit 3aa91c009017b508c0f6298f7e8ee3a622bdccc2
  Author: Katsumi Yamaoka <yamaoka@jpl.org>
  Date:   Tue Apr 27 06:38:57 2010 +0000

      Don't load tm and apel XEmacs packages when compiling.


It looks like the key code is in mml.el in gnus, but I haven't
understood it yet.

Is perhaps the issue that the multipart/signed code is called without
the forwarded heder visible, and then the forwarded header is also
changed to the same value.