One of the problems with a more realistic test is that the injury risk increases dramatically. Isn’t it better to have a simplified test that even though not completely realistic gives a good idea of what you would get in a realistic situation and keeps the risk at a level where you can get participants? I for one would not participate in a test where you evacuate in to a burning field.
Since all tests have approximately the same error the result should be valid. Then you can argue if it is enough to bring everyone out in 90 seconds using half the doors. My best guess is that for situations where it isn’t enough a lower time would not make a real difference.