No, I had CISCO on the phone.
It's an issue related to the MAC address controls used to prevent a single user node from stealing the entire DHCP scope, the adapter bonding used to fuel DOCSIS 3.0, and what happens when you have two 1:1 NAT enabled clients a little too physically close together trying to talk to each other a little too much.
I've gone over all this before, so I don't blame COX, it's definitely a CISCO thing. Just to further reinforce this, I have a buddy of mine that recently graduated college and started in internship in CISCO's Edinburgh office. He was rather interested in the troubleshooting process I was going through on this issue, so I kept him in the loop and sent him all my notes. He asked a question about it, being an IOS intern, he had access to the code in question.
The next I saw the guy he was still trying to recover from the lashing they gave him. Apparently they thought he was some kind of plant. He's under NDA so he won't say much but he did confirm it's a "bug" in the current implementation. But it's one of those design things that's always worked and now doesn't in very specific conditions. So I doubt we're going to see a fix soon, it sounds like something akin to asking Untangle to stop using Java.