It's not the bandwidth, it's the latency.
Normally you'd see a 95% hitrate in the D$ and I$, 98% hitrate for L1+L2 caches. This gives you an average memory latency ~3-5 cycles. When you switch core, every memory access causes a miss and have to spend >20 cycles to get to the caches on the other core to get the data.
In CS:S I see 20% lower performance in in-game situations, and it's the anoying kind, not just average lower framerate, but "hickups".
The fact that I didn't reinstall XP when I dropped my X2 in my existing motherboard might have something to do with.
Cheers