An official position is that one should run multiple Redis servers per machine and scale by splitting datasets between them

Which doesn’t necessarily help if you have certain shards that are hit heavier and it’s more maintenance.

Also, if you are running on prem, you can optimize your VM for cores/RAM. At least on AWS, you can’t perfectly optimize your core count versus RAM vs speed.

But then again, if I were running on prem, I might use Redis because it handles use cases by itself that it would take Memcached+ other servers to manage. With cloud hosting, I could use managed purpose built services if I needed more features than Memcached can handle by itself.