A Rice CS team of Sandler, Mislove, Post, and Druschel have sketched a scheme for distributing webfeed content using P2P publish and subscribe, called FeedTree. To appear in the 4th International Worksop on Peer-to-Peer Systems, FeedTree is designed to help deal with webfeed bandwidth issues. I call it a sketch because the paper is a bit short on implementation details and has no performance analysis, but as Wes Felter says, "maybe Rice will actually ship".
The tricky bit, which the paper tries to address somewhat but not satisfactorily, is how to deal with the legacy implementations. You could be looking at an IPV4 vs IPV6 situation again. Also, deploying to PlanetLab seems lame to me. Build something real publishers might want to use and go from there, even if it's a limited subset of the real world. Hack up a budding, open source, client like Jaeger and work with one of the webfeed engines (PubSub, Technorati, Bloglines, FeedBurner) to see how this could play out. That'll provide bigger impact plus an independent, empirical analysis of webfeed behavior would be more useful than simulations on PlanetLab. Besides, you'll need a model for those simulations anyway.