In recent years, satellite networks have emerged as a critical access medium for global connectivity. New Low Earth Orbit (LEO) satellite constellations have enabled Internet access to remote regions where deploying terrestrial radio access networks would be economically or technically infeasible. Despite these advancements, constellation networks continue to face persistent challenges, particularly in reducing end-to-end latency and increasing available bandwidth to support seamless communication. Smart strategic placement of Multi-access Edge Computing (MEC) resources in space or on Earth could play a critical role in meeting these performance demands. This paper proposes a comprehensive analysis of Round-Trip Time (RTT) for computation services under various edge server placement strategies, including space, terrestrial, and hybrid cloud configurations. We use MeteorNet, a novel real-time emulation platform that integrates orbital propagation models and networking virtualization, to evaluate realistic satellite constellations and flight-grade application behavior at system-level scale. Our formulation of the edge server placement problem as a non-linear integer program enables quantifiable comparison of configurations, subject to constraints imposed by dynamic topologies and infrastructure limits. Through experimentation, we identify conditions under which terrestrial edge clouds provide optimal latency and cases where space-based architectures outperform their ground-based or even hybrid counterparts. We also demonstrate how server placement strategies, such as cross-plane, along-orbit, and farthest-point sampling, affect latency performance, as reflected in the average routing path length within the constellation. These findings provide actionable insights for the design and deployment of MEC server placement in space-based cloud architectures, demonstrating the value of real-time emulation platforms, such as MeteorNet, in supporting future edge computing platforms.