Of course it also protects against server failure, but only to a certain degree. If the master server fails an application can continue to work with offline data but without the master server to broadcast change notifications there is no real-time collaboration anymore. And the risk that the offline modifications lead to non-trivial conflicts at the time the master server comes back and the changes are to be merged up increases over time.
Leveraging the same replication mechanisms that CDO uses for offline clone repositories you can easily build up fail-over repository clusters. The most sophisticated topology involves a separate fail-over monitor which manages a dynamically growing or shrinking set of repositories. It elects new master repositories in case the current master fails and communicates with special fail-over sessions in the clients.
With the new CDO Offline Example application it's now easier than ever to play with these complex setups. I've recorded another screencast (which builds upon this one) for you to give you a clue on how to start:
Switch to HD on YouTube
I hope you're inspired by the demo and develop new ideas for your own applications...