Good point; I agree this makes code much more naturally machine graded than essays, and a statistical classifier for source code sounds disturbing (although it might do better on grading style than the usual autograding does).