401 and beyond on p5b leaves much to be desired, at least in terms of the mbench synthetic. take for example this run at 500 1:1 4-4-4-4-25-3-9-4-10 (compare it to the 400 4:5 500 one)

Access latency 60.0 ns (240 clocks)

Read datarate (INT) 6797 Mb/s
Write datarate (INT) 2773 Mb/s

Read datarate (MMX) 7421 Mb/s
Write datarate (MMX) 2767 Mb/s

Read datarate (SSE) 7821 Mb/s
Write datarate (SSE) 9022 Mb/s

hmmmm