I've switched from using Thunderbird to offlineimap and mutt. My mailboxes have over 50k messages. After tweaking some settings, mutt is very responsive with this many messages. notmuch is extremely fast for searches. But offlineimap can take several minutes to sync all the folders.
After a little research, it appears that IMAP was not intended for use with very large folders. If that's the case, then what kind of emal workflow should I be using to keep folders smaller?
Not all IMAP servers are fast, but as previously noted, it is possible to have an efficient IMAP implementation and setup. I've had reasonable luck with dovecot. However, offlineimap is a particularly slow program. It worked fine in small tests, so I switched over to it and redid my whole email setup to depend on it. But then the more mail I got the slower it got, until in the end I started losing my mind every time I checked email.
If you want a completely offline search experience and like notmuch, you might consider muchsync, which is kind of like the equivalent of offlineimap but specialized for notmuch to take advantage of notmuch's index. muchsync is much much faster than offlineimap once you'd done the initial sync, because it avoids all the network round trips that offlineimap has and takes advantage of notmuch's index data structures for efficiency. I switched over to it from offlineimap and have been very happy.
Yet another alternative to consider is isync. I have no personal experience with isync, but some people seem to like it. Having previously used and hated offlineimap, I'd be seriously investigating isync if I hadn't switched to muchsync.
There is the off-chance that you are referring to uwimapd, an implementation for an IMAP server, when you say "IMAP". In that case you need to make sure none of your mboxes becomes larger than 2GB, that is all emails in the same folder, not counting sub-folders.
If you are not referring to uwimapd, I believe your problem is the performance of your computer and internet uplink, not IMAP. Therefore:
Only keep emails no older than 3 months and those belonging to ongoing projects. Put the rest in an archive folder. Create the same folder structure under archive as you have under INBOX. That makes it easy to find stuff in archive, if you can't find it in INBOX.
IMAP, the clear text protocol your mail user agent employs to retrieve and manage the emails stored on the (remote) mail server, becomes problematic when it has to handle thousands of emails at once. The key aspect is "has to handle". The overall amount and size of emails is not a problem for IMAP, per se. The problem arises if it has to actualy perform work on a lot of emails.
This happens, for instance, if thousands of emails have their content and/or status (flagged, un-/read, priority, timestamps, ...) changed between two syncs. As you use offlineimap this may indeed happen, depending on your use case. However, in that case the only thing that will help you are more frequent syncs.
"After a little research, it appears that IMAP was not intended for use with very large folders."
My opinion (and experience) is exactly the opposite. Imap has very sophisticated server-side searching, indexing abilities - the only problem is, that most mailclient uses this practically as a remote file handling protocol.
For example, with IMAP is it possible to search for mails in a folder with a given SMTP header. Or you can handle the file attachments of a mail separately from its body. But this has the price: the protocol is much complexer as any other I've ever seen.
The real power of the imap will be only visible, if
the actual IMAP server is able to index your mails
and your mailclient is able to handle the imap folders intelligently (i.e. not using the imapd as a remote filesystem, but doing most of it with remote queries on the server side).
My only good experience was the thunderbird/dovecot pair with it. On the clientside, kmail is also relative okay, and also cyrus on the server side. Another systems aren't really okay.
In thunderbird you can make "search on the server" queries and see their results as a virtual folder. And the serverside search allows the imapd to make fast, keyword-based searches based on its internal index data.
I don't know offlineimap, but I know mutt. Clean, char console apps are very good and mostly very efficient, with a single exception and exactly this is what you found: they tend to avoid the indexed things and try to things linearly (although very fast).
I've just added imapsync to Alpine Linux. It can be run in an LXC container on any Linux. An Alpine container is around 5mb in size.
I transferred a 7GB mailbox to a new mailserver last night without any problems. It took about 3 hours with 2.2 messages / second transferred & the process used 260-280mb of RAM. It takes 60 seconds to sync 5 mailboxes with --useuid caching enabled.
