To the contrary, we have had quite good results in virtualized environments and you don't really need timing that is that accurate to make it work.
If you don't handle RTP, I'm sure it is amazing. However, if you have to do voicemail, stream audio from the server or do any kind of actual time/latency/jitter sensitive processing, I don't care how much you tune your hypervisor, it's never going to scale.
We work quite well on amazon EC2 for example. There are 2 issues I know about with vmware, 1 is you need to set a setting on the host to extend somewhat sane clocks being available, the second is I have seen issues with the bridged network adapter actually doubling up all packets causing very strange issues, I suggest not using bridged networking if you experience this.
I've not seen this behavior on Vmware ESX 3.5u2. Maybe an issue on Vmware Server or Workstation?