If you have routing loops in your network, chances are you have bigger issues to worry about..
Discontiguous networks create a supportability problem, and increase the risk of these very issues. One is not well advised (IMHO) to create such a network without a darn good reason...and this (to me) does not seem to be a Darn Good Reason.
Oh, and the mentioned "stardust" problem will happen whether you choose to deploy distributed mail/dns/cache servers, or whether you have a big farm. You'll still need some form of load-balancing in front of your cluster.
That's exactly the point. It is easier to solve this problem (not to mention scale or plan) for a farm or cluster than for servers in each of 270 pops. And easier yet when I use names, rather than the silly "well knowns". Opinions aside, the infrastructure required to support stateful load balancing for these boxes in each of their routing islands is substantially more than most are committing these days... Regards, Mr. Stardust tns-inc.com