If you changed UTF-16 to UTF-32 or UCS-4 I'd support it. I think there are already implementations that use the replacement character for all "impossible" codes.