Sounds fairly arbitrary, assuming that's actually a quote from a "this is how we define 'abnormal' regarding this medical/psychological issue" kind of official document.
For a more mathematically-based arbitrary limit-bounding, I'd treat the areas below and above the 1st and 3rd quartiles as "not normal[2]", on the basis that a solid half of the population are in the bit not worth talking about because they aren't extraordinary in either direction (for this single measure, of course, see my prior point), but whether it's 40% or 50% deviation from population norm (and whether that's in sample-size or axial deviations), then there's likely to be very little difference between a person 1% within the limit and a person 1% without it. Certainly the measure involved is likely to be swamped by all kind of other individual characteristics not involved in the measure concerned.
Sorry, I'm not sure if this is a point worth making. This thread's a little deviated, itself, from its original aims. (And it's doesn't seem worthy to make a joke along the lines of the people likely to contribute to this thread being anal over such small details.)
[2] Further sub-dividing the area (not deviation!) in half again if you really need some sort of "abnormal" equivalent term for the furthest reaches of each near-asymptotic bell-bottoming curve, rather than "just outside the norm".