It would be just as fast if it detected the locale correctly, and used the UTF8 state machine for the UTF8 locale only. Other locales could have their own state machines or just fall back to a generic implementation.