Not to say that hosting a personal FriendFeed and attempting to link it to my friends and their own FriendFeeds does not sound like fun, but it would be mind-numbingly tedious for most. The friction to get started is far too high for this project to ever gain real traction.
He goes on to make several variations of this argument. I think he overestimates the complexity and cost of getting servers up and running on the internet. He also underestimates the ability that people have to automate the process.
Depending on how the Diaspora project is done, it might be possible to host it on a regular web hosting site that supports PHP or some other scripting language. It could also probably be wedged into a Google Appspot instance. VMware and or Amazon EC2 could also come into play.
For any given popular web platform, eventually a number of providers of that service arise and will take care of the details for you. I'd imagine that Robert Scoble is already talking to his coworkers at Rackspace about making it happen for Diaspora once they get all the details.
If the guys get it done, I expect to be able to rent an instance for about $5/month, if not less. (10 instances in a family plan for $10/month wouldn't be too much of a stretch). For commercial free social networking, it would be well worth it.
I could also imagine the larger ISPs bundling it in, or making special provisions for it in their traffic management, because local ip traffic costs them far less than packets that traverse the backbones.
I hope that Diaspora and other projects get off the ground. It'll be good to open up the web again.