On 9/3/07, John S Russell
Because the circumstance that Joe is proposing here is one where your TCP/IP connectivity between (C) and (A), even with the usual internet tricksy mutiple-path, dynamically routed, cleverclogs stuff going on, is still broken.
Maybe (A)'s route advertisments have flapped a bit recently and been damped. Maybe a couple of big ISPs associated with that network have chosen to de-peer from everyone else and now you can only reach it via an accoustically coupled modem in a goat shed/Data Center in Badbusinessmodelstan. I'm told that sort of thing does actually happen.
But the other MX on (B) is still reachable from (C) despite any or all of these possible reasons because it's on a completely different network and probably (if Joe's being sensible - and he always is[1]) one that's been chosen specifcally to be as diverse as possible from (A).
This is a pretty common setup for services like MX's, DNS servers, and other stuff that you'd like to have reachable more often than not, where support for multiple servers is built into the protocol itself.
That's all fine and nice. What I'm saying is that the MTA in [B] should have almost the same configuration as the one in [A]. It should certainly be able to reject e-mails for non existent users during the SMTP session otherwise the MTA in [B] can be a source of never ending NDRs. Bojan