I'm looking at the stale-* options again and wondering if we have any good reason to port the update_headers option that is part of that feature? It provides an HTTP violation.
I think the disk->memory promotion we already do for HITs deprecates this as a performance boost. Amos
