As requested I have redone the calculation that was here but now assuming S,T,U,V,W,X,Y,Z are no longer normally distributed but rather are the lengths of the intervals separating sorted random uniform variables in [0,1]. This also is known as the Dirichlet model.
Under this new probabilistic model:
Instead of 19.6%, my Monte Carlo program finds
as the probability P of a favorite betrayal scenario in a 3-canddt IRV election: