The JXTA site's tutorials section includes a really nice 140-page PDF tutorial for people just getting started on the subject: Project JXTA 2.0: Java Programmers Guide.
Before beginning the column, I had no idea that JXTA was language agnostic. The only thing it has to do with Java is the fact that the reference implementation -- a 528-class, 88-package monster -- is written in the language. JXTA's a group of related XML dialects that could be implemented in C, Perl, UserTalk, or favorite language here. Peers don't have to be running the same implementation -- my Java-based peer could form a group with your peer written in Kvikkalkul.
The subject gives me an idea which is probably being implemented somewhere: At some point, the popularity of site syndication formats such as RSS is going to hit a scaling wall, because it's inefficient for thousands of people to request a feed 14-16 times a day just to check for changes.
I think a cloud of JXTA peers could be the solution: In a group of peers, ask if anyone has a copy of a particular feed stored within the last hour. If they do, request it from them instead of the publisher. Otherwise, go get one yourself and cache it for others. A client-side aggregator such as Radio UserLand or NetNewsWire could form a peer-to-peer network of its users or a larger network of any RSS-consuming software users could be created.
You can see fledgling support for such a scheme implemented in one of the Mac RSS
readers, namely Shrook (http://www.fondantfancies.com/shrook/), where its refered to
as distributed checking (http://www.fondantfancies.com/shrook/distfaq.php).
Must admit, I'm a NetNewsWire user, but Shrook is innovating fast, including a synchronisation component ...
Though the project is dead, Reptile's ( http://reptile.openprivacy.org) goal was for a p2p aggregator.
The Bouillon project may be of some interest to you. It is an universal reputation-driven P2P aggregator (current implementation, JBouillon, is made in Java).
Currently in the stage of public testing.