Exoskeletons are human-robot interfaces that have enormous potential to assist people with everyday tasks. To improve the design of exoskeletons for use in clinical populations, it is important to further our understanding of how exoskeleton design and control parameters lead to sub-optimal effectiveness. Here we simulated the effect of three factors, gait variability, wearer-exoskeleton delays, and exoskeleton inertia, have on the predicted energy assistance provided by an exoskeleton with a finite-state controller trained on a set of stroke survivors' free walking gait data. Results indicate that larger errors between the wearer's desired ankle trajectory and the exo's estimated ankle trajectory result in statistically large reductions in the actual assistance provided. Specifically lags on the order of even 10 ms can illustrate statistically sub-optimal performance. Likewise subjects that exhibit large gait variability will have a statistical reduction in actual assistance. However, reasonably low exoskeleton inertias are not significant as a factor in terms of sensitivity to wearer assistance. Therefore, to improve cooperative control algorithms for exoskeletons and achieve true assistance based on wearer induced motion, this work implies that designers should prioritize minimizing delays and wearers should train to reduce variability in order to maximize energy savings.