If one uses Tor, the article states that the person will not be treated as a United States person unless "proven" otherwise. If that's the case, all the traffic from a Tor exit node will be considered as traffic from non U.S. persons and all data will be stored. Doesn't make sense to me because if you are using Tor right, then requests cannot be traced back to you. How does that help?
The NSA is very good at exploiting very small mistakes. In the Venona project, http://en.wikipedia.org/wiki/Venona, they NSA realized that the soviets were re-using one time pads. By knowing A xor C and B xor C, they were able to get A xor B - and from that, they were able to decode the messages.
I could imagine things like looking at the tor traffic, and trying to match up things like writing patterns. And of course, there's the very good chance some people aren't using tor right.
I don't think that what the NSA is doing is legal - but think there's a good chance it's somewhat effective.