I like the anon comment that they should not re-invent the wheel so much. That gave me a chuckle, I'll just use my off the shelf distributed web app deployment software that runs on thousands of machines ... oh snap!
Great to see this in one place. I would love to know more about the more internal facing systems (code deployment, test frameworks, failover systems...).
It worked well in the past. They made significant improvements to it.
However, it's not as well-suited for multi-datacenter usage. For example, they needed another replication layer. I think they're working on a current system that will replace memcached, and it seems to be causing some consistency problems...