Do you have benchmarks to back that up? The efficiency per core is worse, but I’ve not seen 10x worse. One thing to be careful of is making sure to have parallelism in your benchmark. Single-threaded Postgres may well be 10x faster on workloads of interest, but that’s not very representative of real use.
We switched to CockroachDB for some of our most heavily loaded tables a few years ago - and the scalability and operational simplicity has been great.
I don’t have a like for like benchmark, but at least for writes - we’re running with 5 replicas of each range so that’s 5x the disk write i/o just from that.
Add in some network overhead and running raft etc, and you are going to be needing a lot more resources than for PG to maintain a similar throughput - but adding resources when you can scale horizontally is so much easier that the tradeoff is worth it in my experience.
Though it scales horizontally, you will need to throw at least 10x or more hardware at it to achieve the same throughput