Very true. It's expensive and complex to have real HA/redundancy, but they can't even bounce a DB server without taking down every one of their apps. I have no idea about administering Windows servers, but based on my experience with Windows Update on the desktop, I would imagine that you need to do that relatively often.
Hopefully they have a staging environment running somewhere internally at least.
One also has to take into account how many systems that are distributed over n servers have actually n points of failure. My estimate would be "quite a lot".