What is apparently supposed to happen is that when one ESX host in a VSA cluster fails (either due to network or server hardware) any VM guests that are running on that ESX host power off and reboot on the other ESX host.
This power off is not a graceful shut down, nor can it be I suppose.
What I did as a little tweak to my Windows VM guests was change the start up count down in Windows from 25 seconds to 5. Since the servers do not shut down gracefully when they boot back up the screen comes up that asks you if you want to boot into safemode, start normally, etc. This way you save 20 seconds on the boot up unless you watch the console and press Enter.