It?s also possible that the cache files USED to grow quite large in previous versions and that now they are much more reasonable, but the only way to reset their size appears to be to delete them and let dovecot rebuild them.
I seem to recall these cache files ballooned in size when I did some doveadm search queries -- under some circumstances some headers fetches are cached for later searches, which may explain the growth over time as clients do mailbox searches.
As for Solaris10 mremap_anon() space failures, I'm a litte hazy on memory mapping and shared memory, but I wonder if this has anything to do with kernel maximum shared memory limits. I know that for some of our Solaris 10 servers, I had to up these values in /etc/system to get some of our services to work.
Joseph Tam jtam.home@gmail.com