Heya.
I might be mistaken, since I never owned any of the ATi's previous generation GPU (X300, X600, X800, X850).
But I believe Adaptive Anti-Aliasing is a new A-A method for the X1K series (X1300, X1600, X1800), that allows minimal performance hit when enabled, and provides the same A-A quality, if not better, than regular non-Adaptive A-A.
Also, Adaptive A-A anti-aliases things like fences and a lot of angle-dependent textures (or something similar) much more efficiently than regular A-A.
Xbit Labs have an article about pretty much all of X1K features and technologies, here is the link:
http://www.xbitlabs.com/articles/video/display/radeon-x1800.html
I think they're talking about Adaptive A-A in there, I'm not sure though.
One thing is for sure, is that Adaptive A-A is the best A-A method ATi managed to create so far.
Temporal A-A was introduced with the previous generation though, and if I am not mistaken, the Radeon 9xxx series is capable of it with Catalyst driver series 5 (5.1, 5.2, and so on). But Temporal A-A works well only when V-Sync and/or Trilinear Filtering is enabled (again I'm not sure about that, but I do know something must be activated for Temporal A-A to function properly), and is limited to 3x.
Adaptive A-A is currently limited at 6x, at least on single-GPU configurations. Although 4x Adaptive A-A is enough for most games.
I've tried 6x AAA in many games (using X1800 XL), and I saw no noticeable differences with 4x.
I mean ... when you play a game, everything is in motion, so you can't really tell if there is a difference. But if you stop completely, and take the time to look at borders and walls or your weapon on the screen or stuff like that, then you might see something. But I don't think you'll take the time to do that in a firefight in Counter-Strike, or when your base is under attack in UT 2004, or when you need to take and hold a flag in Battlefield 2.