As IoT applications continue to expand, the demand for computational offloading to handle the vast amounts of data generated by IoT devices is constantly increasing. Cloud computing alone is often insufficient due to high latency and network congestion, leading to the adoption of cloud-edge-IoT architectures that leverage edge servers to process tasks closer to the data source. As an alternative, the transparent-computing rationale decouples data fragments from (executables of) programs and allows migration of either data or code so as to meet each other at some computing device (be it a cloud server, an edge server, or the IoT device generating the data) that will conduct the required computation. This versatile computing paradigm poses a new challenge, that of optimizing the orchestration of computational tasks in such environments, based on real-time performance measurements of the underlying infrastructure. In this work, we formulate this transparent task orchestration problem as a multi-objective optimization problem, aiming to minimize latency, energy consumption, and load imbalance. To solve this problem, we employ the NSGA-II evolutionary algorithm, a widely used heuristic method for multi-objective optimization. In order to boost its performance, we propose a sophisticated initialization strategy based on a \(\epsilon \) -discretization of the solution space, as an alternative to the typical random initialization. Our approach incorporates task dependencies, realistic communication constraints, and server utilization models, making it applicable to real-world cloud-edge-IoT systems.