I've been spending most of my courier effort these days on the x86_64 version... I've gone to current versions of courier and courier-auth and simplified things a bit by going back to exim's smtp auth mechanism and supporting PAM as well as mysql so that "real" user accounts do not have to exist in the database. I've further stopped supporting pop-before-smtp using drac since it doesn't scale much better than the log scrubber and every email client I've seen recently supports smtp auth...
I have only had a few people interested in the script and I haven't gotten much feedback on it, so I probably won't continue developing it for a general release. I am available to convert new DA installs to courier, but I'm not confident enough in the mbox to maildir conversion to support the conversion of populated systems.
I don't expect that you'll see a plugin version anytime soon... The process is simply too complicated.
I've worked with dovecot for awhile, but I think the push from DA to go to dovecot is because it supports the mbox format, so they don't have to convert from mbox to maildir. While I'm sure the performance would be better than UW, the bottleneck will likely still be the size of the mbox file, so I don't think going to dovecot without going to maildir will help the situation much. And if we're going to maildir, I don't think the conversion to dovecot will be any easier than what we're doing for courier.
The new version of courier-authlib helps things a bit on the authentication front. It should be much easier to utilize the virtual user password file for authentication if you don't want to use mysql. However, from the courier-authlib documentation:
userdb is a simple, straightforward solution that scales to a couple of thousand of mail accounts, depending on the hardware. Beyond that, one of database-based modules will need to be used, such as authldap, authmysql, authpgsql.
Essentially, he's saying that using a single flat file for courier authentication should scale to a couple hundred accounts. Using a nested userdb scheme with multiple domain userdb files should scale to a couple of thousand accounts. And if you need more than that, go to a database auth mechanism.
My thinking is that mysql is as critical as email in a shared hosting environment, and in a busy system it will actually be much less expensive to do a simple database lookup than to parse through a flat file for a password. However, if using mysql is the big hold up here, I'll happily look at implementing something that's compatible with DA's current auth mechanism.
On the DA side, however, I think the hangup is still the mbox to maildir conversion. My script uses the mb2md perl script which seems to work relatively well for this, but if we can come up with a rock-solid mbox to maildir conversion it will probably make the conversion to courier an easier feature to support. If someone wants to volunteer on that front, I'll look at the other auth schemes and update my script for i386 and x86_64.