Moved to ePO for better attention.
One possible way which i think might work
- Create a second UNC repository on each ePO server that replicates to it's opposite on a regular basis.
- Setup up the respective repo in the agent policy on each ePO server.
In the event of one ePO server failing the clients will then attempt to contact the secondary UNC repo (on the still working server) instead of the master as it's unreachable.
Once your aware of the failure you can the configure the working ePO server to update/replicate to this UNC repo.
You're not going to get reporting on clients associated with the failed server but they should in theory be up to date.