I remember hearing about China usurping 15% of western internet traffic for 18 minutes. This was accomplished by having nodes report as being the next closest hop in the network path to the packet destination. In a decentralized darknet project, I imagine such an issue being much more widespread. In fact, I would imagine a darknet project would actually play in to the hands of the government. It would be perfectly plausible to infest the darknet with millions of your own nodes reporting as the next best hops thus inserting themselves in the middle of all darknet traffic able to analyze data as it flows through the system. Obviously, a darknet would utilize encryption for traffic but all bets are off when you potentially have a constant man in the middle and no centralized authority on trusts. What's worse, and more to the point of playing into the hands of the government is that a darknet would give them (the government) a concentrated focus area. If I were to categorize the percentage of traffic that was "interesting" for regular internet traffic vs. the percentage of "interesting" traffic of all darknet traffic then I would imagine the darknet having a much higher ratio of noteworthy to junk traffic. If I had a limited amount of resources to invest in analyzing and decoding secure traffic I would obviously point my tools at the most richly dense data source.
Man in the middle attacks may not work as well as you think. While it is possible to create false information using this darknet mechanism, it is possible using things like PGP to ensure that you are getting data from the same node. Of course, you have to figure out how to set it up right from the beginning; if someone messes with you before you know what their key is supposed to be, they can feed you false information.
You might be able to overcome this by using a many-to-many authentication mechanism. But realistically, what could the government hope to accomplish by feeding you false information? (this is the only thing they are likely able to do). Consider:
There is no way to back track traffic and discover who sent the request.
Once you realize a source is feeding false information, you know to never trust that source again.
It is easy to imagine a decentralized rating system for the quality of information provided by various keys on the network. Keep in mind you can't really fake who you are. You are your public key, no one else can publish under your public key but you.
Sort of. The only way to tell if a person made a request is if you control all the nodes that this person is connected to. If you do not control even one node that a person is connected to, there is always the possibility that the request came from this other node, not the real node. This is because the difference between making a request for something on your behalf and making the request on someone else's behalf looks exactly the same as far as another node is concerned.
This is how freenet works. Of course, in freenet there is a time to live associated with each request so it will die eventually if, for example, the searched for item is not present on the network at that time. You could figure out that it's from a particular node by seeing what the time to live is from that node, but small amounts of random variance in time to live values can effectively ensure that both requests don't live forever and that it is suitably difficult to determine the origin of the requesting node.
Now, it is certainly possible with enough concerted effort to find out what a user is doing with some statistical probability that a user is looking at something, but you can rarely be absolutely sure.
I think 100% certainty that a node is the originator of a request is not necessary for most purposes. If you are thinking of a court case, then maybe, but only if there is no corroborative evidence.
And in other situations where people might want to use a darknet (e.g. a repressive regime) a few false positives aren't going to bother anyone concerned.
The good thing about the Darknet plan is that it levels the playing field - Sure, Governments might suddenly have access to influencing the infrastructure the same way that corporations have now - but I would argue that Governments have quite a lot of influence already. The important difference would be that ordinary citizens would have that very same power.
Unless you live in a predominantly authoritarian state (and I guess disagreement about this is what most of these discussions come down to, in the end), keeping government in check by empowering the people with insight and access to its processes is usually what enables a democracy.
Your point about darknet traffic being inherently more interesting is insightful, but I'm not sure where it gets us. The same can be said about TOR (and there have been attacks on this, too, as well as TOR posing the same issue of possibly being 'infected' by too many nodes of a single attacker) and I guess it' really just a technical challenge, in the end.
I'm not sure I understand the benefit of it being easier for more people to run man-in-the-middle attacks on the network. I doubt the government would rely on such a network for sensitive information, so it'd probably end up being citizens spying on other citizens (or worse, e.g. phishing).
Sep Kamvar has done a significant amount of research addressing precisely the problems you describe. The conclusion being that (in non-pathological cases, which would likely include the case you vaguely describe) one can design network architectures using reputation mechanisms such that malicious peers are found and routed around. See http://kamvar.org/publications, specifically "Numerical Algorithms for Personalized Search in Large-Scale Self-Organizing Information Networks" (the first Book) esp. Chapter 9, I highly recommend it.
I think the goal of a darknet is more to prevent censorship than to ensure privacy. As you said, any untrusted nodes could easily reveal a user's location and intent.
I would much prefer a network that piggybacks on the existing infrastructure and poses as innocuous traffic through stenography or encryption.
