Within a priority group, virtual machines are always
recovered in the order specified by the recovery plan. High-priority
virtual machines are recovered serially, so that recovery of a machine
does not begin until its predecessor in the list has either been
recovered (powered on and connected to the network) or has failed to
recover within a specified period. Virtual machines in all other
priority groups are recovered serially per-ESX-host, so that in cases
where a group of machines spans several hosts, the recovery plan can
take advantage of the opportunity for parallelism. During this type of
recovery, machines on a specific ESX host are recovered in the order
specified by the list, but the recovery order of the entire list is
subject to the assignment of virtual machines to hosts. For example, if
the first three normal-priority virtual machines are hosted on one ESX
host and the fourth is hosted on a different ESX host, the fourth
machine in the list may be recovered before the second or third.
Because vCenter limits the number of virtual machines that can be
powered on in a single request, recovery plans cannot power on more
than 20 virtual machines at a time even if there are more than 20 ESX
hosts available.
In your case, are the VMs hosted on more than one of the three ESX boxes?
