Has anyone tried setting up more than one cache implementation for a hibernate application?

It seems as if would be fairly straightforward to write a "composite cache provider" which allows you to configure for each cache region which cache implementation will actually be used.

This would allow you, for example, you could have critical heavily updated data being cached using a transactional clustered cache, while less critical tables could be clustered using JVM-level cacheing with timeouts.

As long as there are no dependencies between cache regions at the implementation level, this seems like it would make a lot of sense for certain application.

Are there any gotchas that anyone knows about preventing this from being done, or any reason why this has not been implemented?