In the VAX bcopy function posted, it would probably be faster if the constant 65535 was replaced by 65532 wherever it occurs, thus causing all memory references to be longword references if the source and destination are longword-aligned.