Well, actually, if you look at the semantics of the "permission"
parameter for external-body, you'll find that there are actually cases
in which the hash would be wrong intentionally, because one use for
external-body is to have a pointer that always points to the
most-current version of the data rather than the data as of the
time-of-sending. But when the permission parameter is set to "read", a
hash value would be meaningful. We would get this for free, however, if
we established a Content-Hash or similar body-part-header. This never
quite got settled for the Proposed Standard version of MIME, but I at
least am willing to see it added as an optional header field in the
Draft Standard version *if* we can reach agreement about it... --
Nathaniel