The first alpha of JBoss Cache 3.0.0 - codenamed Naga - is out and available for download. In a nutshell, it is a truly open source (LGPL) distributed enterprise cache, which is often used as a library to remove data retrieval and calculation bottlenecks, or as a mechanism to share state across a cluster either for failover or load balancing, creating data grids, etc. I have also seen it used as a distributed in-memory object database of sorts, as a fast alternative to a more traditional database when typical DB features and guarantees aren't critical.
It has a new Multi-Version Concurrency Control (MVCC) locking scheme that it is completely lock-free for readers, which makes it very efficient for a read-mostly system like a cache. There's more on the MVCC here.
There's also a new, improved layout of the .xml configuration file, and the entire configuration has been refactored to be more readable, concise and consistent. More on that here