Your understanding is correct. The peer nodes in the cluster maintain replicas of your data according to your configured replication factor and search factor. With 2 indexers, you probably have an RF=2 and an SF=2.
The master contains no actual index data, other than what it needs to coordinate replication between the peers. As of Splunk 5.0, the cluster master is a potential single point of failure for the whole Splunk cluster.