Polymeric membranes are efficient in separating gas mixtures, typically by exploiting the sieving mechanism. What controls the sieve size of a given polymer matrix is unclear, although one line of thought implies that the local cage size, defined by the dynamic motions of the glassy polymer matrix, is the relevant metric. Here, we use coarse-grained molecular dynamics simulations and show that the sieve size is defined by a static cavity size controlled by polymer chain stiffness (a packing-driven metric) combined with the local cage-like motions of the polymer host. The best separation performance for a pair of gases is when this combined metric is roughly half way between the diameters of the gases in question, with the static and dynamic quantities contributing roughly equally. For the various models simulated we find the existence of an upper bound correlation which passes through this optimal point and has a slope expected from the Freeman model, namely , where the d's correspond to the kinetic diameters of the gases in question. Our results thus demonstrate that the relevant free volume size that affects gas transport in these condensed phases is defined by both static and dynamic measures.