RFC 2822 is incorrect when it suggests that References has all of the
message's ancestors. What References actually has is the start of the
thread, followed by the most recent ancestors, ending with the parent.
Ancient ancestors other than the start of the thread are often missing.
(This is explained in various pieces of USENET documentation, and was
pointed out in the discussions that led up to RFC 2822, so there's
really no excuse for RFC 2822 error.)
Threaded readers reconstruct the thread by looking at References in many
messages. This would obviously be necessary even if the naive RFC 2822
model of References were accurate: a message doesn't know anything about
other branches of the same thread, particularly branches that don't even
exist yet. (I have no idea what Costello thinks his proposal is supposed
to accomplish.)
As for multiple parents: This is the only time that I'd recommend
actually putting In-Reply-To into a message. The reader, after seeing
the multiple parents in In-Reply-To, will realize that References is a
mix of ancestor information from those parents. End of problem.
---D. J. Bernstein, Associate Professor, Department of Mathematics,
Statistics, and Computer Science, University of Illinois at Chicago