For message/partial, the problem was solved by requiring the use of only
7bit encoding - see section 184.108.40.206 of draft-ietf-822ext-mime-imt-01.txt;
I believe similar prose existed in RFC 1521.
The reason for the restriction was to avoid "nested encodings", which
would require multiple passes over a message in order to decide whether
it was possible to handle the message or not; this was considered a Bad
Thing by the original MIME group.
At the moment, I'm almost tempted to take one of Stef's ideas and define
an Application/MIME type which can encapsulate anything (includig an 8bit
message/*) and apply a transfer encoding, in order to get out of this bind.
But I don't like this "solution".
8-to-7-bit downgrading needs to be studied!