this seems to be fully supported by VMware nowadays according to the whitepaper at http://www.vmware.com/vmtn/resources/798. but there are some catches that this whitepaper doesn't mention.
the problem arises when you connect to the VC with rdp and start the VIclient on there. when you accidentally click on the console button it locks up. now if that was all it wouldn't be that bad since this is to be expected.
but when this happens, the whole network stack of the VC goes down. it somehow repairs itself (???) and after several minutes a new rdp session can be reestablished. and even that wouldn't be the biggest issue.
the biggest issue is that when VC goes into a consoleloop, HA kicks in and orfans the ESX server that the VC is running on. The ESX server goes down, VM's get started on another server and the whole environment gets mucked up. only after 10 minute when DRS did it's job do thing normalize a bit again.
this whole concept of VC in a VM is a nice idea and if you don't have a physical management server where stuff like commandvieweva is running on to install it on, making it virtual has several nice advantages. but having it killed like this with a wrong click in the interface for me makes it not worth the trouble and i'd invest the 1200,- for a dl320 for a physical box.
does anybody recognize these problems or know how to avoid them? this way i don't think vmware should support it! it's too easy to mess things up.