need guidance when running abacus with openmp

[Satinelamp]

When running a large system like an Au cluster, I'd like to use OpenMP to accelerate the performance, but I tried to set OMP_NUM_THREADS=2 and mpirun -n 32 abacus, which would double the time that needs to finish each electronic step compared with when I use OMP_NUM_THREADS=1 and mpirun -n 64 abacus.

Is there some guidance about how to use OpenMP to improve performance?