It would be interesting to pre-cook vault bundles that are very likely to be interesting for download.

Some "interesting" bundles might be:

• the directory tarball corresponding to the HEAD branch of snapshots
• the directory tarball corresponding to release objects

To make the experiment viable we should probably start incrementally by integrating this into loaders and only doing it for the future focusing on:

1. the HEAD branch at each (future) ingestion
2. the most recent release object (according to its timestamp) observed at each (future) ingestion

(2) might it a previously cooked release tarball, but that's fine, as cooking is idempotent.

Once we are sure this is sustainable we can give start retrofitting, cooking old objects.