Not meaning to start a flame war here, but I'm a bit curious about this as well. Are there benchmarks/tests somewhere demonstrating this? The most relevant info I've heard about this sort of thing is "Dragon NaturallySpeaking is/will be SSE2 optimized"--which isn't all that relevant, as that's one app, and it's for speech-to-text, not voice recognition.
Kelledin
"/join #hackerz. See the Web. DoS interesting people."