Looks great! With A8 pro particle instancing is used so that you can show approx. 30k particles in your demo. Besides, try to pass NULL as function and step each frame with ptr_first and .link.next through all particles, updating lifespan to something >0, e.g. 100, and updating the .xyz position.

I am not 100% sure, but since wait(1) is not allowed in particle functions, I guess they are executed at once in one frame roughly as I described above. So, my tip can also likely be no improvement. I guess the main reason why your performance breaks down is just the casual overdraw.