> One thing that's hard to judge from the outside is the quality and efficiency of that code.

I've had the opportunity to work with a few ex-NSA software engineers. One in particular was among the smartest and most capable colleagues I've ever had. If, as is the impression I gathered, he's at all typical of the talent NSA has at its disposal, I would expect the limiting factors to be the volume and accuracy of the data available, rather than the quality of the software built to analyze it.