> nothing but commodity hardware needs to be paid for,
> and big resources cease to be relevant.
Really. Creating Linux/GNU/X/etc was basically a rewrite project for the first decade, even now most of the work is more improving the existing codebase than inventing major new things. Yes some major new things do happen these days but it isn't the majority of the effort.
Now compare to this proposed 'Internet as it was supposed to be, nuke proof, private, perfected' idea. There aren't even any good theories in the academic world to take and run with. So phase one isn't even something for code monkeys, it if for math geeks, network and game theory nerds, etc. And remember that the final product has to withstand active attack by the almost limitless resources and manpower of nation state actors.
So what is the biggest Free Software organization currently. Does it have the resources to build thousand node test networks and then fund thousands of man hours by 'the best of the best' to trying to break it? If any did is it not reasonable to assume they might would have made such a hardening effort at things like Firefox, Glibc, Apache, BIND, etc. already?
That is the big problem with this notion, it not only requires a new breakthrough in network design it will almost certainly require a level of software reliability in the face of an unprecedented level of active attack that has never been achieved to date. Children break Windows and IE, serious Black Hats break LAMP servers. Is there anything the NSA couldn't break if they were desperate? Or the Chinese intelligence agencies? If the final proposed system can't, with a high confidence backed with hard maths, claim to be resistant to such determined attackers then it isn't worth a damn.
To deploy a system as is under discussion that isn't secure will only get a lot of people killed when they foolishly rely upon it and at the critical juncture the secret police round them all up. So either the required miracle in design has to also be a design that can be 100% private, untracable and yet verifiable regardless of implementation bugs or compromised communications links or the fielded implmementation has to prove 100% reliable when the day comes.... and it will probably be put to the ultimate test at most one or two times.