Charles Lindsey writes:
Yes, if having two References headers turns out not to be feasible,
then having one References header plus some number of secondary
References headers with some different name (I don't think I would
want to use See-Also for the purpose) would be another way out.
Sounds like RFC 2822 to me. IIRC 2822 says something like this:
1. Generate a References field using the usual algorithm.
2. If the new message is a reply to multiple messages, copy their
message-ids to the In-Reply-To field.
But all the reports so far (not a representtive sample, to be sure)
indicate that no harm arises from multiple References headers (i.e.
they progagate, and are interpreted as well as could be hoped for).
(The software on your own machine is an exception, in that it didn't
propagate the field.)