For mime-sniffing, doesn't that any content→type heuristic run into the problem that languages are sets of strings, and a string can be in many sets.
This is the same problem as charset sniffing that led to UTF-7 attacks.
That said, a standardized, predictable heuristic is much better.