Who said anything about scaling? I run simple low utilization services on a replica set because of high availability. Replication only provides secondary reads in terms of scaling and that's not the point. Actual scaling comes from sharding (partitioning of data).