L. M. Orchard braindumped his thinking on building a foundation for webfeed driven applications. The key things that come out are 1) feed crawling, 2) archiving data model + querying, and 3) client API. He focuses on the second point and similar to work I did with Jeff Cousens, an MS student, comes down on the side of a light mix of metadata in the traditional relational model, and stashing the raw data. As Jeff showed, this can scale to some pretty large feed collections on stock hardware.
Well, if you consider 1 TB disk stock hardware ;-/ which isnt' that farfetched. However, the key is to store as much raw data as possible. If you get your data model wrong you can always go back and reprocess it.