One word - FINALLY! I've been waiting for this for ages. Thomas Morgner blogged about this a while back, but it was all lies :)
Until now, and here's the proof:
This may look like a cyclic dependency... using a CDA datasource in PRD, while CDA itself uses PRD libs. Believe me, it's not, and it's incredibly handy to have a "dictionary of querys" on the server and just query them from your report. Caching comes for free, it's all executed in the server, it'll save hours of pain waiting for the hard queries to get executed. I know this is something that PRD 3.8 fixed, but having the cache on server is a huge bonus
In order to use this, just drop this jars in PRD-3.8 lib dir and it'll just work!
Have fun - I know we will!