The only thing that comes to mind is like a Mercury Software...Load runner and Win runner...??
I would take this approach:
Have a VM template with benchmark suite installed and some performance counters like CPU/memory, disk I/O, network I/O turned on.
The benchmark suite you use can be the TSScale scripts or pinpoint
You can customize these scripts for your typical office worker scenario.
Create multiple clones of this VM and power them on and have the benchmark script to run at power on. (use the registry Run or RunOnce registry key)
Generate perfom logs from with in each VM as well as reports by the ESX server for memory and CPU utlization, disk I/O and observe the load with different number of VMs powered on.
Another good utility is autoit. It will run the desktop use case for you.
Whatever you do in this area, remember that if you publish anything you are violating VMWare's EULA. I have to wonder why VMWare won't let you publish any benchmarks on their VM's?