The problem with that thinking is that the replication mattered. I'd argue it didn't, it was essentially a scam that people fell for. Who cares about failover when you're losing data due to a bad implementation? Who cares about replication when you can gain the same performance by using a performant database on a single node?

Did mongodb truly allow anyone to really horizontally scale? Most places that need massive horizontal scaling using something like mysql as far as I know.