The most obvious use for this setup (the reason I made several customers implement it at my previous life as an abusecritter) ) is to close down an open SMTP relay that couldn't otherwise be closed down (*cough* Cc:Mail *cough*). Relaying is controlled on the publically accessable server, but only mail destined for the target domain comes into the primary MX. Hence, no thrid-party relaying. -Chris
Are you sure this couldn't be intentional?
I've once seen a setup where you had the lowest-priority MX (by that, I mean the one with the lowest number, in case my wording is ambiguous or contradictory) being some host with an RFC 1918 IP, and then there was a higher-priority MX which was their NAT box. I'm guessing (I never sent mail there, or worked with this setup, thank god) that the idea was that connections to the RFC 1918 box would die, so remote MTAs would contact the NAT box and deliver there. The NAT box would then try to relay to the primary MX, and since it would obviously have an interface into the network with the RFC 1918 IPs, it would be able to deliver. This place doesn't seem to be using this setup anymore, although amusingly enough most of their NS records point to machines with 10.200 IPs.
I agree that this type of thing is entirely dumb, but is there any reason that the network mentioned by the original poster couldn't be doing the same thing? Many large corporations that have been running IP networks since before Wall Street knew the meaning of the word Internet have different real blocks of IP space (usually in the class B space) for their "public" network and their corporate network...
-- --------------------------- Christopher A. Woodfield rekoil@semihuman.com PGP Public Key: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0xB887618B