Chris is correct to watch the machine itself; which is no different than with the execution of any load test. As soon as the machine generating the load becomes a bottleneck (CPU/Memory choking) the results of the test are compromised.
You might want to keep an eye on Throughput, as well. I've seen situations where CPU and Memory on the load generator may look acceptable because they are behind he "bottleneck" caused by the NIC (not enough capacity or even incorrect settings).
I'd be interesting in your findings with VMWare - would you be willing to share what you learn? [img]/images/graemlins/wink.gif[/img]
Well it seems that you are only testing the connectivity. I would think that running this on a VM would be absolutely fine. as long as you don't confuse this with a proper setup to run actual tests then you are fine.
Generating load means you want all the available resources on call. Moving to VMs mean you'll be giving up some capacity, so, I wouldn't recommend it. Let alone, if there are issues, you have both the VM image, and the VM host to troubleshoot to solve.