I don't know about best practice guides (would be nice to see one that's oriented to archives/libraries), but I've had good luck using btmakemetafile which comes with the canonical bittorrent client. It's a command line tool that takes a file or directory as the first argument (the payload of the torrent), and a tracker URL as the second. The key here is the tracker URL, which is used to enable peers to find each other.
A lot of torrent sharing websites essentially act as trackers, in addition to being a catalog of available torrents. You can run your own tracker if you want using bttrack, or you can install a more friendly tracker for managing your collection. Another option is to use an open tracker, or to figure out how to create trackerless torrents (they use a distributed hash table behind the scenes). I'm not sure how to coerece btmakemetafile into making a trackerless torrent though, but I've heard it is possible depending on the version.
I really like the terasaur's idea of a curated collection of torrents for the library/archives community. It would be nice if the process for getting a collection hosted on there was a little bit more transparent, if it enabled a community of curators, not just torrent seeders. A nice side effect of uploading content to Internet Archive and Amazon S3 is that you get a torrent tracker for free. The Internet Archive torrent files can be seen on one of the detail pages.