I don't see that there's a clash. There's a certain redundancy, but
that's nothing new.
If you're using MIME encapsulation techniques to attach an ftp file
to a body part, you use the message/external-body mechanism. But
if you're writing HTML, you put a URL in an anchor.
It might be nice if WWW clients could parse MIME multipart and
message types, but it seems reasonable to just pass them on to
metamail or some such MIME user agent.
On the other hand, a MIME user agent need not grok HTML -- it just
passes it along to the WWW client.
Am I missing something?