does adobe audition 2.0 do phase cancellations?

Dammit!!!  Steve, I spent all night trying to prove you wrong.  Seriously, it seems like if you can subtract C, you should be able to isolate it.  My brain hurts.

Hmm... I think that you should have realised half way through that you'd just ended up with the result that I did... and that the rest follows inevitably! 

Well, I actually toyed with doing this once on the premise that isolating L & R was possible, a premise I mistakenly believed in because I had done some erroneous math.  It's not that I really disbelieved you, Steve, but it was fun to run into the brick wall of logic a few times just to see where I might have gone wrong before.

Yes, intuitively it seems that you should be able to do this - but you have to remember that the centre signal isn't 'real' in any sense - it has no tangible existence of its own. The fact that you can cancel it means nothing, because the only possible way of doing this by direct phase-inversion methods inevitably mixes the L & R out of phase, and we all know that you can't unmix signals... don't we?

The Center Channel Extractor is more devious though, because it treats the virtual image directly. It's not perfect, because it has to autocorrelate it out of two signals that have content that can be widely varying. But nevertheless it can do quite a good job of elimination - sometimes better if you do it in two passes, not taking out quite so much in each.