I did spot that zeroing the upper 32 bits of RAX leads to undesired non-NOP behavior, but failed to convince our team in 2001 to use one of the freed up 1-byte opcodes (a la UDB).
