I took a brief peek at the paper cited by the article (http://theory.stanford.edu/~aiken/publications/papers/sigmod...) to see how the system worked. It uses hashes of n-grams of the submitted programs to compare similarity.

My own experience with plagiarism came when teaching a university course for CS majors. For most, the class was their second course in programming. While grading an early programming assignment I noticed a program that reminded me of one of the early ones I had already graded. Comparing the two programs revealed that they were identical, except that the variable names had been changed to protect the guilty.

I wonder if this system would have detected this case of collaboration.