If our servers are now running multiple processors with their own cache memories and RAID arrays of hard drives, how much more work would it really be to allow multiple servers to link together and run shards of one giant server operating system? Then if one server fails, you swap it out, necessary data and processes are replicated onto it and life continues. You don't need to know which physical machine is running what particular service or storing specific data because the server-room OS handles that. Google, I gather, runs their data centres this way. It would be a lot like running an old mainframe, I guess, but a bit more durable because the components are all redundant, backing each other up and taking over when another one fails. That distinction is important for what I'm going to suggest next.
Imagine doing the same thing as in that hypothetical server room, but with your own computers, and over the internet. Not centralised at some data processing plant owned by a software giant, but communicating and coordinating peer-to-peer online, backing up and replicating your data automatically and turning your work machines and home machines into one smooth environment. That's my dream. I have a lot to learn before I can begin making it real, or else I need to find a lot of people who know what they're doing. I also need to keep fighting against the urge, trained into me by my industry, to centralise everything to maintain control and make life easier. This is not meant to be easy for the ones who develop the platform. They're meant to handle the hardest bits so that users get one seamless environment across all their machines.
Mokalus of Borg
PS - The ultimate vision is a slightly larger job than rewriting Windows.
PPS - So that's probably not going to happen.