Just rolled out our 22,500 account Zimbra system.
(more details if/when I wake up!)
We went live for a College in the US late Sunday night, after approx. 48 hours of downtime from their old system (mostly to apply last minute changes and unforeseen side effects of other requests).
The whole system is behind 2 F5 big-ip load balancers, which balance between 2 servers running Zimbra proxy. Mail comes in to two Ironport devices and is then passed out to 4 mailhosts. We also have 2 Zimbra LDAP servers for redundancy and a Sympa listserv.
Storage is a SAN fabric for all mailhosts and the backup host, storing over 1.5TB of email. We're using a combination of SAS and SATA disks.
All servers are Dell poweredge 1950s with either single or double quad core Xeons, and either 8 or 16GB of ram depending on use. All are running 64bit CentOS 5.2.
Right now we're running about 3000 concurrent users with the majority using webmail.
It's running quite nicely