He says most setups are network or memory bound, so I don't understand this reasoning:

"Well, that’s the plan: I/O threading is not going to happen in Redis AFAIK, because after much consideration I think it’s a lot of complexity without a good reason. Many Redis setups are network or memory bound actually. Additionally I really believe in a share-nothing setup, so the way I want to scale Redis is by improving the support for multiple Redis instances to be executed in the same host, especially via Redis Cluster."

The RAM would have to be a multiple of the dataset, because its shared-nothing, running right into one of the two common limits, right?

I would assume a cluster on a single node would partition the data, rather than replicate it. Without an in-depth knowledge of redis, I can't really predict if 4 instances sized for 1GB would have more overhead than 1 instance of 4GB, but I'm willing to hazard a guess that it would be fairly close. On the network side, it's a wash if you're already bound by the interface line rate.

This reminds me a bit of the K/kdb approach of just starting another process for almost certainly the same reasons.