For our first release we were unable to get VMX 128 SIMD support into the runtime. This is something that we are proactively investigating for our next release.
aL
Have you guys looked in to the Accelerator framework at all
(done by MSR – there was a video on Channel 9 recently). It's a pretty
generalized solution to SIMD stuff written entirely in managed code, and I
doubt it would be too hard to extend it to use XNA in addition to its current
CPU and GPU (via MDX1.1) back ends. You’d get support on Windows and the 360.