Phil Windley, PhD and Terry Wilcox have published a paper, "Evaluation of ESX Server Under CPU Intensive Workloads".  This 32 page paper explores how configuration changes affect the throughput of virtual machines hosted on ESX Server during CPU intensive workloads.  In particular, how Hyper-Threading, Virtual SMP and the amount of RAM allocated affect throughput.

We present a summary of our evaluation of VMWare ESX Server 2.5.2. In particular we confirm and work around known timing issues with guest operating systems running on ESX server. Our work validates and adds to the work of other groups modeling the behavior of ESX Server during CPU intensive workloads by exploring in more detail the effects of Hyper-Threading and the overhead of Virtual SMP. We report and measure a previously unknown performance penalty for allocating too much RAM in virtual machines with Linux as the guest operating system. This paper also describes the testbed we used to manage and run our tests including a virtualization test management system we developed to run the tests we performed. We describe timing issues that affect performance testing on ESX Server and a method for measuring runtimes that gives accurate results.

...

In this paper we explore how configuration changes affect the throughput of virtual machines during CPU intensive workloads. In particular we explore how Hyper-Threading, Virtual SMP, and the amount of RAM allocated affect throughput.

In our work we will use VMware ESX Server 2.5.2 as the VMM.

In summary:

• Single CPU virtual machines scale better than virtual machines using Virtual SMP.
• Hyper-Threading increases throughput if there are a large number of virtual CPUs, but makes no difference if the number of virtual CPUs is less than or equal to the number of physical CPUs.
• Do not allocate excessive resources to virtual machines. Additional resources may hurt performance.

The full detail of the tests performed and the results can be found within the paper, here.