In my experience, SSE is faster than 3DNow on my Athlon XP 1700+ (overclocked to 1.8GHz). Evidently there is some little bug with SSE and some Athlons, so the 3.24 client will pick 3DNow instead of SSE when -forceasm is used, thus slowing the crunching. I use the 3.25 beta client, which fixes this issue (my Athlon has never shown this SSE "bug," so I'd rather have the extra speed).