> When you get down to it, between cache coherency across CPUs and memory, disk flush delays and disk caches, every database is eventually consistent.

This is a false. None of the things you listed preclude consistency.

> And if you want to operate over a distributed network, which means you WILL have network partitions, then you are subject to CAP and will need eventual consistency mechanisms.

That's not how CAP works. Plenty of distributed CP databases exist.

spanner is strongly consistent

Cloud Bigtable is eventually consistent, Spanner is strongly consistent.