Redis eats million of entries for breakfast, is pleasant to work with, has TTL expiration of keys built in and is available as a managed AWS ElastiCache service when you get into serious data sizes: up to 32 core 237 GiB nodes, and then you shard to add more. Redis is also super as a local cache, and simple to deploy and manage together with the app that uses it.

Since you obviously ran some quick benchmarks and concluded that running it locally over a unix socket (confused why you would mention "time needed on the network"... you tested with local sockets, right?) was too slow, you should at least let Antirez know you've run into a new mysterious performance bug ;) Writing a cache service can be a fun side project, but I doubt you gained anything by doing so except another homegrown part to maintain.