Clustering only does three things
1. It copies configuration changes to all units in the cluster
2. It allows you to see the message log as one complete log for all units in the cluster
3. If you are using quarantine it keeps a copy of the users quarantined messages on two different units in the cluster.
If mail is being processed on one unit and it dies then the session fails. The sender should see that as a deferral and should retry the mail at which point it should hit another unit in the cluster.
Clustering is not a failover service. All units in the cluster should accept and process mail as an individual unit. They do not share resources.