Jon,
I'm an engineer in VMware's monitor group, who's done a fair amount of work with VT support. Our initial experience with VT has been that we were unable, for most non-trivial workloads, to eke out gains over our traditional software-only monitor. Since our traditional monitor works pretty well for 32-bit, we didn't put ANY effort at all into performance for our 32-bit VT implementation. That's why it's hidden behind a config option.
So, if you want to benchmark our VT stuff, you've absolutely got to do it with 64-bit workloads. E.g., get your hands on an HP wx4300 workstation, which has a VT-enabled pentium 4-64.
Also note that trivial benchmarks (stuff like getpid, fork/wait etc.) are pretty easy to show a "cheat" on and show an apparent win for VT. More complex workloads are harder.
