Martin, am transferring this to you for adjudication.  I'm -0 on the patch.

On the plus side, the code is well thought-out and reasonably well
tested.  On the minus side, it complexifies the compiler in a way that
may make it more difficult to maintain.

The benefits to the end-user are almost invisible.  

Developers may possibly benefit from the -r command-line argument as a
way of detecting errors in their own code, but this functionality may be
better suited to lint utilities such as pylint which already supports
detection of unreachable code).

We've rejected previous, less sophisticated patches for dead code
elimination on the theory that there was too little pay-off. And at one
point, Guido expressed a disinclination to any efforts to neaten-up the
generated bytecode (he compared it to rearranging assembly language and
thought there was little point to it).