In this work, we compare the wave and loads statistics for two different sea states with a TP = 15.0[s] on a h = 33.0[m] depth, one with a 10-year return period (HS = 7.5[m]) and one with a 100-year (HS = 9.5[m]). For each sea state, a unidirectional and a multi-directional wave realization was measured experimentally and then reproduced numerically via a fully-nonlinear potential solver. The computed wave kinematics were used to calculate loads on a stiff cylinder with a diameter of D = 7.0[m], and compared with experiments. To perform a quantitative analysis, we extracted 30-minute maxima of the free surface elevation and in-line force, and fitted a Gumbel distribution via a Bayesian methodology. The analysis of the experiments showed that the extreme forcing on a stiff cylinder was larger in the 2D sea state than in the 3D sea state. As for the crest statistics, the 2D were higher than the 3D for the milder storm, while they were quite similar for the stronger storm, likely a consequence of the increased wave breaking, limiting the maximum achievable wave crests. The reproduction of the sea states and associated loads via a fully-nonlinear potential solver was overall able to predict the main trends. However, the 3D wave crests were overestimated for the milder sea state, probably due to a too soft breaking filter. The 2D forces for the larger sea state were on the other hand underestimated, likely due to the lack of a slamming load model. The analysis of the average wave shape leading to the extreme load events showed that in the experiments the extreme events are dominated by physics linked with the particle velocity, and hence in phase with the wave elevation signal, as drag loads, slamming loads and velocity-dependent free-surface intersection loads. On the other hand, in the simulations they are more inertia dominated, hence in phase with the kinematic acceleration signal.