I've always assumed that the average properties model is complicated for little, if any, improvement of accuracy(?).
A 50% big / 50% small model is probably adequate and within the relatively wide modelling tolerances. More importantly, it has the advantage of correctly reporting the stress at the small end? I came across a calc recently where the the small end didn't show an overstress with notional average properties but was overstressed with correct properties !!!! ?