This work aims to investigate methods to effectively design human-robot teams using computational working and simulation paired with a human-in-the-loop experiment (HITL). This HITL relates human-robot teaming fluency metrics (human idle time, robot idle time, and concurrent activity) to perceived workload. We also present a computational simulation that is able to predict these fluency metrics. We modeled an on-orbit maintenance scenario where participants interacted with either a robot capable of fetching and inspecting or just inspecting. To mimic a real space scenario, we also tested the differences between having participants listen to all mission control commands and confirmations versus omitting this dialogue. We found that participants rated a robot capable of fetching as a significantly better teammate even if this caused a slower overall scenario with less work for themselves. Robot idle time and concurrent activity were good predictors of how participants perceived the robot. However, more human idle time did not always correlate to lower perceived workload. Computational predictions of the relative effects different work allocations were confirmed by the experiment. Further, bootstrapping analysis demonstrated how the computational models can be further improved from HITL results, both in terms of refining estimates of specific activities and in terms of identifying other important effects to incorporate, such as communication times and the time to transition between actions.