IO is often still the bottleneck. Although the bandwidth is larger, main memory access latency is still high, that's why AMD's 3D cache can enable performance improvement by just providing more cache.