It's a framework & methodology for benchmarking virtualization platforms against one another. It's also a great learning tool to see the differences introduced by configuration changes (such as No# VMs per LUN configuration, networking) For example we're using it to benchmark every type of host that comes in the door to build up our database. We see some interesting changes of the results based on storage configuration (particularly). If you wish to publish any results publically, there are pretty stringent rules regarding the configuration you used to benchmark.
Theoretically I see no reason why VMMARK could not be used to benchmark different hypervisors.
/kimono/
