Scott Beaulier notes the correlation between legible scripts and quality of written answer and puzzles as to the reason for it.
There is, of course, a simple explanation. If you are pretty sure you have the right answer, you have a strong incentive to make sure the guy grading it knows what you're saying. If you're pretty sure that you have no clue, you have a strong incentive to make sure the guy grading it has no clue what you're saying so you can pray for part marks from grader exasperation with trying to puzzle out the content.
Same in papers you might be refereeing. Obscurantism for the sake of obscurantism, or hiding the lack of a plausible mechanism in a mess of math, serves the same function.