I think it's appropriate to consider that this particular test ("driver-side small overlap") is not a random situation to which the car simply isn't specificall

I think it's appropriate to consider that this particular test ("driver-side small overlap") is not a random situation to which the car simply isn't specifically optimized for, but rather a de facto adversarial example - the particular model of accident was chosen because people observed that this type of accident seems to cause a lot of injuries, i.e. that this test intentionally targeted a specific known weakness of the current cars; so it's no wonder that the initial test results confirmed that. If the average car would already (pre-modifications) do well in driver-side small overlap situations, then this type of crash would simply not get chosen for extra testing.