Standalone bytecode-modifying tools almost never check that they're
outputting correct bytecode. http://code.activestate.com/recipes/277940/
makes no attempt to check what version of Python it's running under;
running it under Unladen Swallow 2009Q1 would have produced completely
incorrect code because we had modified how opcode arguments were stored.

I don't want to encourage people to write tools that operate over
CPython bytecode. As such code proliferates, the alternate
implementations have to chase it down and ask that it be removed. It
complicates the process of verifying that other implementations are
compatible with CPython.

Like I said earlier in this issue, for me this is more a compiler
cleanliness issue rather than a performance issue.