We undertake the first study of two-planet microlensing models recovered from simulations of microlensing events generated by realistic multi-planet systems in which 292 planetary events including 16 two-planet events were detected from 6690 simulated light curves. We find that when two planets are recovered, their parameters are usually close to those of the two planets in the system most responsible for the perturbations. However, in one of the 16 examples, the apparent mass of both detected planets was more than doubled by the unmodeled influence of a third, massive planet. This fraction is larger than, but statistically consistent with, the roughly 1.5% rate of serious mass errors due to unmodeled planetary companions for the 274 cases from the same simulation in which a single planet is recovered. We conjecture that an analogous effect due to unmodeled stellar companions may occur more frequently. For seven out of 23 cases in which two planets in the system would have been detected separately, only one planet was recovered because the perturbations due to the two planets had similar forms. This is a small fraction (7/274) of all recovered single-planet models, but almost a third of all events that might plausibly have led to two-planet models. Still, in these cases, the recovered planet tends to have parameters similar to one of the two real planets most responsible for the anomaly.

We conduct the first microlensing simulation in the context of planet formation model. The planet population is taken from the Ida & Lin core accretion model for $0.3M_\odot$ stars. With $6690$ microlensing events, we find for a simplified Korea Microlensing Telescopes Network (KMTNet) the fraction of planetary events is $2.9\%$ , out of which $5.5\%$ show multiple-planet signatures. The number of super-Earths, super-Neptunes and super-Jupiters detected are expected to be almost equal. Our simulation shows that high-magnification events and massive planets are favored by planet detections, which is consistent with previous expectation. However, we notice that extremely high-magnification events are less sensitive to planets, which is possibly because the 10 min sampling of KMTNet is not intensive enough to capture the subtle anomalies that occur near the peak. This suggests that while KMTNet observations can be systematically analyzed without reference to any follow-up data, follow-up observations will be essential in extracting the full science potential of very high-magnification events. The uniformly high-cadence observations expected for KMTNet also result in $\sim 55\%$ of all detected planets being non-caustic-crossing, and more low-mass planets even down to Mars-mass being detected via planetary caustics. We also find that the distributions of orbital inclinations and planet mass ratios in multiple-planet events agree with the intrinsic distributions.