Ok. Heres how were are bustin it. We have 4 mail stores each with the capacity for 250,000 users. We need availability, so each mailstore is a cluster - but here is the funky business. We only have 5 physical machines for the stores.
All 5 machines are are san attached and utilize a clustered file system (GFS). On top of the clustered file system we have 8 virtual machines - zim1-1, zim1-2, zim2-1 and so on. Each virtual machine is a clustered service on the physical cluster so live migration is possible of any of the VMs to any of the 5 physical machines.
zim1-1 & zim1-2 are clustered together and run a clustered zimbra service mail1
zim1-2 & zim2-2 are also clustered together and run clustered zimbra service mail2
and so on.
Each mail store has it's own set of luns on the SAN for index,logs,redologs,db/data,store,secondary that are exposed to the physical machines and are exposed to the VMs as physical devices.
In front of the mail stores we have proxy/MTAs.
The bottom line is we saved a shed load on having to only purchase 5 machines instead of 8 for the cluster, yet have the same availability and more management. It is, administratively, a bit complex - but it is working.
If anyone is interested in this setup we will be putting together a white paper in the very near future.