Yeah, they did roll out the new memory manager, I believe in Catalyst 13.3 beta, IIRC. I could swear I've written about this, but can't find my own source to cite.
The improvements weren't that dramatic, and when I asked about it, the AMD rep told me they'd realized that the memory manager alone wouldn't be a universal cure for high frame latencies. He emphasized that AMD has changed its tuning process to look for trouble spots where there are rendering slowdowns, though. I believe the revamped memory manager can help with tuning once problems are found. The bigger change is that AMD's process now includes frame latency awareness.
You can see they are optimizing for smoothness now in our more recent results.
For what it's worth, it's also apparent that AMD doesn't care about eliminating some of those "heartbeat pattern" spikes that show up in Fraps, so long as they don't disrupt animation smoothness. Nvidia tends to optimize away even those, which is probably not entirely necessary but may pay dividends in scenarios where performance is just marginally good enough.
I think there are further opportunities from both sides, extending down into GPU hardware and driver architecture and especially power management schemes, now that folks have locked on to the proper optimization target, which is consistently low frame latencies, not FPS averages. You can imagine a power-management scheme that saves up TDP "credits" when latencies are acceptable and spends them (by clocking up) when frame latencies start to rise. Excited to see how the next few years play out.