I don't consider something backed up unless there are at least 3 copies of something in at least 2 locations. Also, replicated dbs are not backups, but may be the best available option when you have too much data to reliably backup en masse.

I've done replication on write scenarios. Generally when I've setup mongo or elasticsearch for searching/read performance, I'll also push to a fixed JSON+GZ on S3 or similar. It tends to work pretty well as a fallback for larger data scenarios as a fallback. Had to use it once, and was so glad to have it.