You can browse collections and then order by seeders. Here are some collections I think should be mirrored:

http://academictorrents.com/collection/deep-learning?sort_fi...

http://academictorrents.com/collection/medical?sort_field=se...

http://academictorrents.com/collection/joes-recommended-mirr...

I'm assuming those globe icons means there's a web seed (HTTP server) involved? If so, I'd prioritise the ones that don't have that backing.

Yes. But webseeds also go away over time when a department shuts down a server or a student account expires.

It would be great to get a personalized RSS feed of torrents. This way, knowledge of at-risk torrents can be distributed efficiently. It even allows prioritization.

I wonder if you can set up a distributed-hash-table sort of thing that lets you reliably query for less-popular torrents. Like a magnet-link system that supports top-k queries.

Sometimes if a torrent doesn't have seeds it is a signal that people do not value that data. Not all data should be stored forever. The community should vote with their servers what we should store forever and at what level of redundancy.