In distributed electric propulsion (DEP) arrays, complex flow phenomena and strong interaction between the internal flow of the engines and the external aerodynamics demands the use of advanced numerical methods. Separated and unsteady flow during transition and high angle of attack flight, as well as off-design conditions such as engine failures and wind gusts, make scale-resolving methods like Large Eddy Simulation (LES) the most suitable modelling approach. A GPU-accelerated wall-modelled LES (WMLES) solver was successfully employed for eVTOL aerodynamic analysis. Previous work focused on external aerodynamic phenomena, such as transition from hover to cruise flight and hover in ground effect, using a volume source term to model the engines. However, to accurately characterise safety-critical flight conditions, such as the effect of engine failures on adjacent engines or the performance penalty caused by steady wind or gusts during hover, the whole aircraft and multiple engines with resolved turbomachinery components need to be simulated. Since numerous flight conditions need to be analysed with high accuracy to characterise the aerodynamics of an aircraft with DEP, it is key to maximise the computational performance of the solver. In the present work, the main bottlenecks of the solver were identified and removed, which resulted in improvements in the dynamic load balancing, voxelization and the update of the signed distance field. A detailed comparison of the computational performance on four simulation setups shows up to 86% runtime reduction. An eVTOL aircraft is simulated in two off-design tailwind conditions, for which the flowfield and engine performance is analysed. Finally, a comparison with wind tunnel measurements is presented. Results confirm that GPU-accelerated WMLES is a suitable approach to simulate DEP arrays both regarding accuracy and computational cost.