I created muscle-driven simulations of split-belt treadmill gait that minimize muscle fatigue and metabolic cost. What actually drives human adaptation to split-belt treadmills is up for debate - is the nervous system reducing asymmetry, or simply minimizing effort? Theoretically, humans should be able to harvest mechanical energy from a split-belt treadmill by taking longer steps on the fast belt. I simulated the case where behavior is driven purely by reducing effort with a musculoskeletal model, and found that the model does choose significantly longer steps on the fast belt. Much longer than humans do on real split belt treadmills, in fact, suggesting that even if effort minimization does drive split belt treadmill adaptation, it is counteracted or constrained by other factors.