The ideal adsorbed solution (IAS) theory is the benchmark for the prediction of mixed-gas adsorption equilibria from pure-component isotherms. In this work, we use atomistic grand canonical Monte Carlo simulations to test the effects of molecular siting and adsorbent energetic heterogeneity on the applicability of the IAS theory. Pure-component isotherms generated by atomistic simulation are used to predict binary isobaric isotherms using the IAS theory. These predicted isotherms are compared with those obtained by a full atomistic simulation of the binary mixture. Binary mixtures of argon, methane, and CF4 in silicalite are found to obey IAS theory, while benzene/methane and cyclohexane/methane in silicalite are nonideal. The mixture of argon and CF4 is ideal despite the large difference in the sizes of the two species. This contradicts previous hypotheses in the literature, which state that mixtures of species of unequal size do not adsorb ideally. The nonideal behavior of the benzene/methane and cyclohexane/methane systems occurs because of adsorbent heterogeneity in these systems, which depends on both sorbent and sorbate. In addition, we use a lattice gas model with parameters derived from atomistic simulation to demonstrate analytically that a sufficiently energetically heterogeneous adsorbent will result in the breakdown of IAS theory even in the absence of interactions between sorbates.