Never seen socket hang up on any production instance, even kf/kc/ee.kobotoolbox.org, which until a couple weeks ago all ran on a single, 4-CPU VM.
You can distribute Enketo (and KoBoCAT and KPI) across multiple front-end machines, or simply have these Node.js and Django applications run on a single, separate machine, while the database servers (Redis, PostgreSQL, MongoDB) run on another.
Unknown Error occurred during attempt to update cache sounds like Enketo’s having a problem connecting to the Redis cache instance. Is it just app.js working hard? Are there no Redis issues, like high CPU consumption by redis-server? Are there as many app.jses as your double-digit CPU cores, and are they exceeding the maximum number of open files or maximum number of connections allowed by Redis? Are your disks fast / is there a lot of iowait?
This is more Linux administration stuff than KoBo stuff 
Good luck!
