There shouldn't be any limit on the amount of VMs that can be running concurrently - at least there are none on ESXi5.x, but I can be wrong with ESX 4.
The place to look at are the log files which can be observed via an SSH session to the ESXi as a root. The logs to look for are the following in /var/log:
/var/log/vmkernel <- the log of VMkernel itself
/var/log/vmkwarning <- the warnings extracted from VMkernel are dumped there
/vmfs/volumes/<vmdatastoreUUID>/<vmname>/vmware.log <- the VM's log
To best way to observe what could be going wrong in the vmkernel is monitoring it in "real time" is to enter
tail -f /var/log/vmkernel
and trying to power on the 20th machine by yourself - there you will see the initialization process by itself.
Also, try investigating the 19th Machine if its VMware tools booted up - this is the safest measure for the boot-up chain to continue.