Did not research it for Maxwell, because I generally solve one static Maxwell model per 10000 HFSS simulations...
But in HFSS from v.13 to v.15, in models with magnetic materials used, I had optimal performance growth up to 6 cores. At 8 cores the improvement was marginal. At 12 cores (Dual-processor) there was aalready a negative affect of slowing down, probably because of progressively growing CPU-RAM I/O overhead. 16 cores per task was totally counterproductive.
After v.15 the DrivenModal solver was updated and recompiled. Per-core performance was boosted with new CPU instructions, and possibility to offload the work to GPU (for pure dielectrics only).
But the multicore performance today is AWFUL. Practically, HFSS main solver does not use more than 2 cores for computation in v.2020, and v.2021.
I will not surprise, if situation is the same in Maxwell.

Benefit of multicore/HPC today is only in quick variations sweep, not available in scripting.

Yes, Ansys simply shot its own foot making purchasing of additional HPC licenses pointless to general designers.