I posted this today on our internal blog, it may be of some interest to you.
vSphere 5 (ESXi) has a known weird problem on system boot when using software enabled iSCSI (eg NOT using hardware iSCSI HBA’s). ESXi will check each and every iSCSI connection it knows about (so every path to every iSCSI object) up to 9 times during the boot up process. This happens regardless of the status of the connections – disconnected adapters that have had a path assigned to them will be checked just the same as active adapters. The result is that ESXi will appear to hang near the end of the boot-up process on the iSCSI initialization then the path_mask initialization. The server will eventually come online and will operate normally but the boot process can take a very long time to complete.
VMware suggests as a workaround that as few paths as possible be used between the host and the iSCSI targets which seems to defeat the whole purpose of having redundant multiple pathways between the hosts and the iSCSI targets. At some point a patch will be issued to correct the problem. For now the GA release of ESXi 5 and the first patch released for it exhibit this behaviour. If this behaviour cannot be tolerated then stick to ESXi 4.1 for the time being.
The VMware KB article on this is here: http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2007108