Does this mean they don't actually use Redis, but have an alternate implementation of the protocol? In such a case, wouldn't it be different it terms of performance when compared to the original one. Also, keeping up with new versions should be a major overhead. If Google is still doing an alternate implementation, I wonder what could be the reason.
It generally scales and performs better than normal tech stacks, but its components are so different and so mutually interdependent that it's bad at directly running normal software. The protocol, however, can be implemented just fine.
To be clear: Even though I worked at Google in the past, I don't now, am not speaking for Google, and know nothing secret about this product. I'm extrapolating from widely known general knowledge about Google's technical infrastructure.
I would say yes, it's still different than using the original one, but that's the way it is when you use any managed service.
If yes, that's 5% redis features.
Hardware, networking, routing, load balancing, memory and disk IO, number of instances, backups/failovers etc. are going to be deciding overall performance rather than the Redis binary itself.