This depends on how much hardware you want to waste as "spare resource", and also in how many ESX servers you are willing to loose in case of failure. You may need to try to find the fine tunning on the number of hosts in an HA cluster that fits your needs, i.e. 5 hosts, 1 configured as Host failures. You can have several HA clusters with no problem.
I usually leave the value in 1, or a %, and do a fine tunning on which VMs I want to be failed over and which ones are not important, so they will not be "wasting" spare resources.
Hope this helps.