Partitions happen, often due to human error. Asymmetric communication scenarios are not exotic at all. It's very common to see them when some link is saturated in one direction. This requires no complex routing, and I've seen it from something as simple as a badly configured backup script.
Also, beware of confusing your local failure detector for a global one. The emergent behavior is a pain when you don't take this into account.
Actually, we're not really an EC2 competitor. In fact, in the future we're just as likely to be an EC2 consumer.
What we provide, today, is a premium service that can take a small amount of Ruby expertise and make it much more effective.
What we will provide, tomorrow, is the above, plus an infrastructure for distributing your applications into the cloud(s) of your choice. That's Vertebra (disclosure: I designed Vertebra, so I may be biased).
Amazon has everything to gain from our efforts. We aim to make Ruby, Vertebra, and our customers into first-class citizens in the world of EC2 and beyond.
Also, beware of confusing your local failure detector for a global one. The emergent behavior is a pain when you don't take this into account.