This problem may seem out of place in this forum, but I think the problem is unique to XP VMs hosted in ESX. The same issue was not observed in a similar setup running in XP hosted in ESXi.
I have an XP guest with multiple NICs configured. One NIC is connected to a 10.4.0.0 network witha 255.254.0.0 mask, meaning it expects to handle all 10.4.X.X and 10.5.X.X traffic. However I have another NIC that is connected to 10.5.217.0 network with a 255.255.255.0 mask. XP handles this without problem. I can access 10.5.217.X traffic on the correct interface and all the other traffic back on the 10.4.X.X and 10.5.X.X traffic on the correct interface.
The problem comes up when I try to access another network on the 10.5.217.X interface. The other network is 10.5.195.0. I add a static route to 10.5.195.0 to use the default gateway address of 10.5.217.1. This does not work because XP has chosen to use the 10.4.0.0 interface, even though I can ping the 10.5.217.1 router directly. I try again and this time I specify the 10.5.217.X interface and now I can access the 10.5.195.0 network through the correct interface. This will work until the XP guest is rebooted. After that, the XP guest reverts to using the 10.4.X.X interface for the 10.5.195.0 network and that doesn't work.
In ESXi, a reboot of the XP guest didn't cause the routing to revert to another interface, whereas in ESX I'm seeing this behaviour.