The problem with this is the Wechsler model (like other IQ models) is (in theory) normally distributed, but 3d6 is not (its kurtosis (-0.4045) is off compared to a standard normal distribution with kurt 0). So it has too data much in the middle compared to the tails.
View attachment 156781
If you wanted to generate ability scores which were (very close) to normally distributed, you would have to roll something like 15d2-12 (kurt = -0.133, so closer but still not ideal).
15d2 - 12 has mean 10.5, SD 1.94, skew 0, and kurt -0.133 and is the best I have found so far that most accurately models a normal distribution.