A more effective approach to addressing the issue is to create a test/example case. Advanced users can then evaluate and potentially offer solutions
It would be helpful if you could provide details on how you obtain and process the input images, as well as the OCR execution method (API, wrapper, executable). Examining this could reveal opportunities for speed improvements, particularly by minimizing IO operations.
It's worth noting that there have been reported problems with OpenMP on Linux and Mac in the context of extensive OCR tasks, as outlined in these GitHub issues: [1], [2].
Investigating these and other performance related) issues may offer insights into potential optimizations.