Whatever about the comparison to PS2 for relative bandwidth, the benefits are still there to allow upwards of relatively large data structures with high access patterns to be available on demand near to the chip instead of waiting on going to main memory and dealing with whatever extra considerations there are for contention. MS clearly designed this with the intention of mitigating this potential issue with its inclusion of additional prefetching options and added logic (in their custom hardware scheduler which has not been discussed in detail yet, likely because of currently non-optimal software defining its behavior) and extra data move units to help that happen more seamlessly and with less burden placed upon the GPU itself. Looking just at the local memory isn't enough when they put all of these pieces in there to facilitate some sort of ideal operation based on future expectations of its use through the lifetime of the hardware. I am not stating that it will be the secret sauce that puts it over its competition, I am stating that the clear philosophy of the system is to offload as much traditional, general tasks and its burden off what it very weak tablet/laptop class hardware, specwise, so that they're getting more useful output for the duration of its active development. They put a lot of fixed function hardware in there to help that happen.
