It can bring a cost savings, but the savings is very slight if you're comparing apples-to-apples. For example, if you're using RDS in a multi-AZ replication set up with periodic S3 backups you have to take into account running each of the replica instances and the S3 costs. What we found was that RDS was treated like a black box to our automation (which was a good thing for a long time) and so we were less attentive to being over-provisioned and actually testing failover situations than we should have been.