I have added the very first Optix (and RTX hardware trough Optix) support. This work is ongoing on optix_intersection branch and will be merged with main branch after the v2.4 release.
Note: The following test shave been done on a NVIDIA RTX 2070 Super.
Simple scenes
As expected, it provide very little or no benefits for very simple scenes:
PATHOCL with CUDA => samples/sec 7.29M
PATHOCL with CUDA+Optix => samples/sec 7.35M
Optix speedup: 2%
Medium scenes
On a 4 millions (3932.2K) triangles scene but with a very simple shapes (i.e. basically a cube and a sphere):
PATHOCL with CUDA => samples/sec 4.60M
PATHOCL with CUDA+Optix => samples/sec 5.42M
Optix speedup: 18%
Complex scenes
On a 10 millions (9456.6K) triangles scene with a quite intricate geometry were LuxCore BVH has never worked very well:
PATHOCL with CUDA => samples/sec 0.91M
PATHOCL with CUDA+Optix => samples/sec 5.50M
Optix speedup: 604%
Conclusion
Optix/RTX speed up will be highly related to the type of scene rendered. Do not expect any particular speed up on (geometry) simple scenes. On real world complex scenes, the benefit will start to be noticeable. There will be corner cases where the speedup will be insane ... still laughing about the 604% speedup

Short version: it works as expected.