PhysX runs slower on a faster machine

PhysX version: 3.3.2
Platform: Linux

Machine #1 (laptop) CPU: Intel Core i7-4720HQ
Machine #2 (desktop) CPU: Intel Core i7-4930K
Machine #1 g++ version: 4.9.2
Machine #2 g++ version: 4.8.4

Both have 16GB of RAM.

The PhysX setup consists of a single scene with a static rigid body actor and a cloth being draped over it, all within 100 frames/iterations (i.e. 100 calls to simulate()).
There is no rendering being done, just pure physics simulation.

On machine #1, the run time is around 15 seconds and on machine #2 the run time is around 17 seconds.

The run time is so long because the meshes are high resolution (i.e. 1000s of faces/primitives each).

I was really shocked to find that the faster machine performed worse than the slower machine.

I’m guessing that the g++ version difference plays a part, but I would have thought the superior clock speed would have made most of the difference.

Can anyone shed some light on why this might be happening?