VMware: ARP request to VM only works from other VMs, not from outside

[ATC]

We have a VMware ESX 2.5.1 server running multiple guests. One guest running CentOS 5.2 intermittently becomes unreachable to all hosts except the other guests running on the same ESX server.

Investigating the issue, we found that the guest did not receive the ARP who-has requests originating from the physical network, and thus did not send any replies. If a ping was initiated from the guest to a host on the outside, the proper ARP entry would appear in the outside host for a few seconds but then expire and once again the guest would become unreachable from that host.

[ATC]

In "/etc/sysconfig/network-scripts/ifcfg-eth0" on the CentOS guest, the following line was commented out:
"HWADDR=00:0C:29:70:2C:31"

We then did a
"service network restart"

...and the problem vanished. I can't really explain why, because the hardware address matches the address of the virtual network adapter, and is the same address that appeared (and still appears) in ARP tables on the other hosts.

I would love to hear from you if you experiences the same problem, if this solution works or not. Any input would be greatly appreciated -- floyd@atc.no