Distribution means that some entries are not available locally and hence will need to query other nodes. How much more cheap/expensive this is compared to a round-trip to the database is the type of question you should be asking yourself. It also depends on number of cluster nodes. Replication works well up to 4-6 nodes. Test the performance of your use case with replication vs distribution, that might give you a better idea which one suits better your use case. With replication, you can work on the local data without problems because all nodes contain the same data (total replication, all data replicated to all nodes).