I'm clearly not pcwalton@ but my thoughts would be: It would be a cool experiment to verify the Mio interface.
i.e. If an entirely isolated re-implementation of async-io in Rust can benefit from or easily standardize to a common interface, that gives the interface far more validity as a good/simple/generic interface.