I'm basing it on practicality. Step 1 is designed to measure a students general medical knowledge as determined that someone should have after 2 years of medical school. The specifics of the test, however, are variable in both space and time. What this means is that if you took the exam an infinite number of times and we plotted your results the distribution would be Gaussian. This shows how your 3 digit score is not very useful anyways and would be better represented as a range, or confidence interval if you're more inclined to stats and probability theory. So how should we compare people?
My point is that it's very complex when you start talking about doing stats with these scores so I think it's best to follow Occam's razor and simply approximate it with statistics based on a normal distribution, since we all can understand that, and I don't see a more accurate method.
Therefore I disagree that a 250 being 84% vs 90% is significant. The practical purpose of the score is to 1)see if you pass and 2)to compare you to others for residency. Basically if you get a 250 it doesn't matter (practically) if its 84 or 90, if you don't get your specialty of choice it was for other reasons.