Nathaniel,
I share your concern about the base64-only vs. more-general usage of
checksums. On the other hand, there seems to be a clear consensus that
having it on Base64 is a Very Good Thing. Most folks seem to think that
it should be required (and I'm biased in that direction) though that
clearly is not unanimous.
Late-entry ideas always are dangerous, but this one seems too good and
too well supported to let slip by.
I think that putting it on the -Encoding header is 'just right.' I
suggest defining it for all -Encoding headers, but declaring it required
for base64. (No, I don't really think that the suggestion is very good,
but I can't think of anything better. In particular, I think that
the checksum for text-oriented encodings will be worse than
useless, given the liberties that relays take with text.)
Dave