It's only randomized in the sense that it pulls from all categories that you include on the test design screen. But, the catch is, it runs out of the minor subjects pretty fast (when I was 2/3 done it had run out of nearly half of the subject categories). So then it won't pull from those subjects.
I agree, somewhat of a weakness. Either this is a design flaw in the bank or they purposely did this so that at the end of the bank (if done randomized) you are only left with the big subjects and no longer messing around with "minor" subjects like biostats, molecular bio, histo, etc, etc.