I've often wondered: what's preventing someone from building an IM client on top of e-mail? While it's very inefficient for short messages, we're not using dial-up anymore (and I'd argue it wouldn't be that bad even on dial-up). Furthermore, e-mail could serve as only the slow, max-compatibility path, with some fast path (IMAP public folders? Or even a completely unrelated protocol) when the users are using compatible clients.
Okay, so I guess the idea is not so much IM-over-e-mail, but rather a schema that maps IM operations into e-mail operations.