Here are the alerts which we think are important, the rest we turn off:

• CPU % Utilization – various measures – Fatal level only 
  • Cluster CPU Utilized > 60%
  • ESX CPU Utilized > 70%
• VM % ready > 5%
• VM and ESX Memory Swapping – any is bad, not the same as OS level swapping, this is swapping at the ESX host level, causes massive performance issues
• ESX or VM Memory Balloon Deflation Failure (Balloon memory not being released, indicates an OS level issue)
• Cluster vRAM allocated > 100%
• Cluster vRAM Utilized > 70%
• Low Datastore Disk Space - Cluster datastore with less than 50GB free is a real issue when you have to do a restore
• Low VM Volume disk space – Fatal level only – need to decide on thresholds (C:\ Drive with less than (vRAM Amount) space free, for example)
• VMWare Tools not running/reporting - only send alerts once per day
• ESX Hosts disconnected from VirtualCenter
• ESX Host multipathing outage
• Agent Data Collection Alerts - working on some self-healing capabilities for these
• FMS Operational Alerts
• Rule to clear any alert after 3 days - assuming you don't have people actively clearing them (shocking, I know)

I'm looking for others' suggestions for important things to monitor.