HA tries to power on a maximum of 32 VM concurrently on a given host. It doesn't wait until the VMs completely boot up, it's only the "Power On" event which HA waits for to finish until more VMs are going to be powered on. With your example of 64 VMs on a failed host this would mean - assuming you only have one more host in the cluster - HA issues a power on command to 32 VMs, wait for the "Power On" task to completes and then powers on the next 32 VMs.
André