From a purely data-mining perspective, the problem with that is that it is possible to build up a good idea of who the unknown terrorists are by seeing who else these people phone.
The thing that confuses me, though, is that it is surely more efficient to start building the network from a small number of known individuals and construct the network based on those communications alone, rather than adopt a blanket technique as they have done. As the content of the telephone calls are not known, it would seem to be that it would be very difficult to identify new Terrorists from anything other than a connection to known ones anyway.