Actually, a counter-point: there might be situations where you want an anti-cryptographic hash function - where similar inputs produce generally similar hashes.

It could have some application as a preprocessing step in clustering (if it could be done efficiently).