Especially when he said that some people did have to work on that setup to scale it up. Sounds like devops work, so the same would have happened with redis, right?
He covers this - the operational setup & expertise to scale Memcached already existed because Etsy already used it elsewhere. It would have had to be built and learned (mostly) separately for Redis.
Maintenance is also something that has to be learned (and "built up" in the sense that you're going to accumulate playbooks and processes over time).
If a team already existed that knew the ins and outs of memcached, knew how to resolve any problems it encountered, already had appropriate monitoring / alarms / etc, that's a massive leg up over needing to do all that with a completely new technology.