When one application server is sufficient to handle all the voice traffic for a location, adding another application server into a cluster provides redundancy. While both servers are online, the servers share the traffic load. But if one server fails, the other server must handle all the traffic. As long as voice traffic is within the capacity of one server, you have full redundancy and there is no interruption of service and no loss of data.
If the traffic requirements require a cluster of two application servers running at or near capacity and you add a third server, the traffic is spread equally among the three servers. If one of the servers fails, the traffic load is divided between the two remaining servers. As long as the traffic is within the carrying capacity of the remaining two servers, there is no interruption of messaging service and no loss of data. The behavior is the same in a cluster of four servers.