First off, FRA won't catch any terrorists. They don't communicate by msn messenger. They use encrypted messages. If they use i.e. 256bit AES encryption, it'll take 30 years minimum for a 2,5 quad core processor to crack it. FRA will use 14000 xeon processors, but they still won't catch any terrorists. It will take too long time to crack, and too long time to discover.
But if somebody is watching the central exchange point, keys need to be exchanged in encryption initialization...and guess where those keys are going to pass. Take a look at RFC 4432 and you'll see what i mean. I have a home server running BSD UNIX that acts as DHCP/NAT/FW primarilly, eg. acts like router so whole internet traffic from and to my home network goes through that single point. I tested several out-of-the-box secure applications, such as secure-pidgin, https pages and IMAP via SSL email standard. Guess what? Simple packet sniffer running on my BSD box caught all keys, and logged raw encrypted communication, and i descrambled afterwards using those provided keys.
Imagine how many servers are between me and you in this moment, and how many points where you can sniff. That's called Man In The Middle scenario, and the only encryptions immune to this are keyless ones.
Organized terrorist groups have experts. And they probably have programmers that are advanced enough to implement their own custom encryption algorithm. Governments use public encryption methods because they can control all the points of communication. Keyless custom encyrption algorithms are unbreakable (even by brute-force, you don't know the mathematical logic of the algorithm). But they have their drawbacks, and i won't drag this offtopic anymore
In any case, they're going to gain - nada, null, zero, nothing by monitoring internet traffic. They're maybe going to caught some skinhead groups planing the attack on the local China Shop, or football hooligans consolidating attack on the adversary group, etc, or local script kiddies / hacker wannabees doing some web site defacement. Someone with a bit of resources and a bit of adequate brainpower will be totally immune to this.
However, it may bring a piece of the puzzle to add to other, more standard mechanisms (police, army, secret service etc...), in terrorist hunt. However #2, this is the most absurd human rights violation...i'm going to write mails to someone describing how i played Red Alert and "nuked Americans' base", it's going to trigger the system and some jerk over there is going to read my mail, and there could be some deep personal and private things involved also, in the same mail.
It is possible to implement a system based on neural networks that would be able to understand something in it's right context, with gigabytes and gigabytes of raw input material for perceptrons. But, to be able to do that in real-time, for (x) languages, counting in the spelling and grammar errors and all variations such as l33t quasi-language...don't know, but that system would be
huge. I don't know if it would be even out of our league regarding current advancements of computer sciences. That kind of system would be then already used (CERN and FermiLab both ring a bell), yet they don't have anything as powerfull and "intelligent".
So, in my humble, but also expert and in-field opinion, that thing is going to work on pure statistics, trigger words, repetition, that kind of stuff, sorted out by "relevance" depending on that statistical factor. Eg. a dumb word counter damaging your privacy.
Somebody influential has lobbied the law, to use it in his own advantage, of course.