I've been looking for something like this as well for some time, and have been looking closely at OpenTSDB as a "data warehouse" for the time-series data contained in Zenoss' RRDs. OpenTSDB seems ideal for the storage (if not the presentation) of that data, and I haven't seen many other open source alternatives.
As an aside, the storage of time-series data presents a number of challenges to data storage, some of which have been addressed by RRD, Whisper, etc.
Anyways, I have written a ZenPack (still testing myself so definitely YMMV) that acts as a "shim", patching the "writeRRD" method (and others) used to store data in the RRDs.
The idea is that the "writeRRD" method(s) will quickly pop its data to a RabbitMQ queue (same name as the RRD path) with the timestamp and value data that would be written to RRD. A separate Zenoss daemon process then combs through the various queues and pushes the data via socket to a running OpenTSDB database.
It seems to work pretty well (largely), is fast, and is designed not to impact the data collection performance. Also, this Zenpack could be easily modified to populate some other data repository, since the key part of it is getting at the Zenoss data without having to comb through the RRD files periodically.
Curious to hear any feedback, although I have to say i'm definitely no OpenTSDB expert as of yet.
The ZenPack is here:
http://wiki.zenoss.org/ZenPack:OpenTSDB
Regards,
Joseph