I'm curious, as previous research indicated to me that this was a bad idea. I'm not sure what improvements have been made in 5.1, as I'm still assimilating data. I had to upgrade my entire lab to 5.1 this past week, and a "suggestion" was made to just put everything (vmkernel management, vms, etc) on a dvSwitch or series of dvSwitches.
Do the same conditions apply if, say, the entire datacenter goes down (i.e. all ESX hosts and the vCenter VM running on them), where you are basically SOL until you get the vCenter VM moved back onto a standard vSwitch in order to power on and get networking again?
In 4.1, in a previous life, a situation arose where I got a call when an entire blade chassis (the datacenter) went down and a dvSwitch was used for everything. In that situation, vCenter was also running as a VM, and when the datacenter was operational again, I had to go in and move the vCenter VM to a standard vSwitch to get operational.
Forgive the ignorance, I've been out of the game for a little while as my time spent was in other projects, including openstack.