Number of ejections as a function of the effective quality factor of the planet. Each dot is representative of a run of 100 simulations done with the same value of Q. We observe that the number of ejections using the semi-analytic model is greater than using the N-body. The first reason is that the percentages are computed with all simulations considered, smooth evolutions included. The difference then gets more and more pronounced as the number of smooth evolutions also increases with Q for the N-body code (Table 4), but stays rather low for the semi-analytic simulations. The reason for such differences is probably that the rates of mean longitudes are not properly modelled when we average out all the terms except the resonant ones and such bias appears during the resonance where mean longitudes play an important role. Also for chaotic simulations, the averaged equations are truncated to the fourth power, meaning that they are not suited for high eccentricities. Iapetus nearly never survives if its eccentricity grows over 0.25 during the resonance. Therefore the threshold for the semi-analytic code was set to 0.25 for Iapetus to be considered ejected.