So you think that with PCIE 2.0, a 512 wouldn't have been faster?
That depands on the game and how it does use memory. PCIe 2.0 x8 equals PCIe 1.1 x16 in speed, so the more textures you can fit in the local memory the less a game has to load them while gaming. Also it depands how much memory your computer has and if the game is a native 64bit one. While the external memory on the RV770 is still 256bit, internally ATI is still using it's 512bit ringbus system to compensate a 256bit only approach.
With more bandwidth you most of the time can save time while loading new levels, the most stuttering while gaming comes not from the graphics-card itself (except you use a crappy card) but from the OS which starts swapping when the game is using huge amounts of memory to the harddisc and vica versa. The best way to avoid that are 4GB and a native 64bit game which can access more than 2GB for sure (some 64bit games still have the 2GB restriction because the game was not modified to handle more).
More bandwith means you need more layers on the PCB because you need twice as much paths to make it work, the more layers a PCB gets the bigger the chance to screw up and failure of the PCB in the production process.
At a price-point of €129,-/USD 179,- for a HD4850 there is no room for such screw-ups.
best,
Eliot