I'm developing a hibernate based clustered J2EE application and have two major questions:
1. I need to use a clustered cache, and saw that there various options (OSCache, SwarmCache, TreeCache, etc.). Does anyone have info/experience on how they compare in terms of performance, data integrity, ease of use, etc?
2. I didn't quite understand from the documentation how hibernate works together with CMT JTA transactions in a J2EE environment. Can someone please elaborate on this?

Any help would be greatly appriciated

2.
When the JTA Transaction factory is used: session.beginTransaction() try to find a Tx in the appropriate context and use it ; if no tx is found, Hibernate will create one and put it in the context.

If you don't use session.beginTransaction() or second level cache, just let hibernate use a datasource and the tx lifecycle will be managed by the container.

Thanks for the prompt answer Emmanuel.
I have another question regarding cache replication:
Am I correct assuming that for clustered 2nd level cache, objects are replicated entirely using java serialization?
For example, if I have a POJO with 5 properties, and I only update one of them, will the entire object be serialized and sent to other nodes in the cluster, or is Hibernate smart enough to replicate only the modified property?
Help would be appriciated

First, this is the job of the second-level cache provider, not Hibernate. If you are using JBossCache (with the TreeCache backend), the subgraph will be fully replicated. If you are using the TreeCacheAOP backend, you get fine-grained replication. This stuff is however in its early stages (I doubt anyone can provide that functionality right now).