Recoveries would (likely, for now - less likely a few years from now) be more often necessary and slower for most users all else being equal, but it's not that

Recoveries would (likely, for now - less likely a few years from now) be more often necessary and slower for most users all else being equal, but it's not that simple. First, if users have software freedom and are running the system on servers under their own control, they are free to spend as much as they like, exactly how they like, to improve reliability. That's the main point of $2^N analysis: if Alice's Transnational Inc. stands to lose up to $500K on a given day if the service is out, she can buy insurance in terms of back-up servers (and her back-up servers only need to serve her users, not every user of the software, anywhere). Alice is not limited to the same QoS as Bob's Main St. Paint Store. 37signals has to optimize its present and future profitability - not Alice's value received. There's a misalignment of incentives, there. Second, if the software were free and generally deployed in a distributed, decentralized fashion, it's architecture could evolve to be more of a P2P system such that many different servers would all have to go down at once to cause more than a slight inconvenience and performance degradation.