Having just recently gone through the experience of replacing the drive used for my VMWare drive images, I've learned a few things. First thing, a creeping bad cluster problem makes for a poor VMWare drive, thus the move. Second, Native Command Queuing did very noticeable things in making my VM's respond more snappy.
When I first made the drive NCQ wasn't common on all drives yet, though these days most have it. On digging into the I/O subsystem (Linux, so this was a bit easier than Windows) I found that the average queue wait times were markedly lower under similar loads, which is a factor of a couple things one of which is disk latency. A 10K drive would scale further than the 7.2K drive I have. It'll still hit performance bottlenecks at some point, even with NCQ, but it'll take longer.
Which is to say, if you're already experiencing I/O bottlenecks with the VMs you're already running, that 10K drive will help. It'll help even more if it has NCQ and the old drive does not. In my case, on my old 7.2K drive w/o NCQ booting two Vista VM's at the same time was enough to send the average queue wait times through the roof. With NCQ, wait times are elevated but reasonable.
As for OS vs. VM, I'd lean VM. Windows is smart enough to allow I/O contention on one SATA drive to not significantly get in the way of other SATA drives. Also, if you're doing workstationy things with that OS drive, like run Outlook and do your day to day tasks on it, it'd take an unusual workload to make it wiser to put the 10K drive on the OS side rather than the VM side. In my experience VM I/O loads are random enough that it deserves the more expensive kit.