Quite. However, if security is what you're after, you may not want to use the most POPULAR one, but rather the better one. Like I said, downloaders, who perhaps care more about easy access (popularity) than anonymity, are a different matter.

If you're submitting to Wikileaks, you're using SSL right? You should be. Exit node sniffing is not a large problem. Furthermore, Wikileaks can just run a number of dedicated local exit nodes.

Leaked documents do not need a low-latency anonymous channel (Tor) to be leaked. Potential leakers should use something like Mixminion (http://mixminion.net/) for high-latency, highly anonymous submissions. Downloads, however, are a bit more tricky, since they DO need to be low-latency.

This is shameless self-promotion, but my colleagues and I have a paper at this year's ACM CCS that addresses just this problem. It's called "Membership-concealing overlay networks," and discusses a network with the explicit security goal of hiding the participants. Since we consider IP addresses to be sufficient to break this concealment, this makes the system also difficult to block at the IP layer. You can find the paper here here, and I would love to get some feedback.

Well, end-of-the-world theories are unfalsifiable. Always bet against the end of the world -- you can't spend the money if you bet on it and win.