I just stumbled across this anomaly while working out some maths to calculate depth of field tables. It's just plain weird. I plotted out all the whole (green), half (yellow), and third (white) stops from 1.4 through 11, as shown in the image below.
If we accept "the rule" as being "round to two significant figures (so no decimal from 10 up) then there are four incorrect values, and two "borrowed' values where a third and a half stop share the same number. Mostly it centres around f/4, where going one step up or down, whether thirds or halves are selected, will get you 0.5 more or less and in three out of four cases that's
wrong. The 5.6 one is the only other and the only whole stop that does not match.
Where it gets even more interesting is if we look at
Wikipedia's "typical" scales (which I am guessing come from the Canikon world?) which also have "errors". Namely, where Pentax rounds the half-stop 3.364 up to 3.5, the "typical" rounds it down to 3.3, neither using the more "correct" 3.4, and also when matching Pentax's "errors" in rounding 3.564 to 3.5 and 5.657 to 5.6.