Temperature-driven floorplaners have been recently proposed to alleviate the thermal problem in 3D multi-processor systems-on-chip (MPSoC). However, the proposed algorithms fail to provide fast placement of the modules when the complexity and the number of functional units in the stack increases. This paper proposes a fast and scalable CPU-GPU implementation of a multi-objective evolutionary algorithm that performs a thermal optimization of complex 3D MPSoCs, capable of obtaining optimal solutions in a reduced time. A comparative study shows that this work outperforms other proposals and reduces the computational time of the thermal optimization of complex architectures.