I like how this makes the loop opcodes more regular - it should make things like Pyjion (http://www.github.com/Microsoft/Pyjion) which translate bytecode into native code.  We already have some code (currently commented out) checking that GET_ITER is followed by FOR_ITER to do some optimizations around loops.  

Another issue we ran into was that this leaves the iterable on the stack while the loop is running...  I'm just curious if your larger change alter that in anyway?