The current cache_storage backend is implemented directly on top of simple disk_cache.  Unfortunately has a couple of limitations that are difficult to overcome.  The big ones are:

1. Its very difficult to implement the VARY header support required by the spec.  See bug 756796.
2. Its difficult to implement the spec'd ignoreSearch feature efficiently with only the single index provided by simple disk_cache.  See bug 682677.

To address this we plan to write a new backend for cache_storage.  This back_end will likely have an sqlite or leveldb database to implement the matching algorithm.  The request, response, and body blob information will likely be stored in simple disk_cache or in flat files.  This still needs to be decided.

Very tentative target milestone of 73.