Yes, hyperthreading splits the cache and gives half to each thread. That's why certain highly optimized code actually performs better with HT off -- if one thread can fill all the Atom's functional units, the other thread adds no performance and reduces available cache. Of course, HT helps in the vast majority of cases, especially with an in-order CPU like the Atom, where it can boost throughput by 30-50%. I know OS X is HT-aware, so perhaps it reports per-thread cache. On my Core 2 Duo Macbook Pro, it reports the full 4 MB L2, but the C2D isn't hyperthreaded.