This is certainly a valid argument. Dwelling into the PhysX Code however is far from a 5 minutes job, as already mentioned above.
One could draw the comparison of buying a car - but all disassembled - and the car dealer says ‘what do you want, it is all there. You got all the source parts’. 
The entire point of using UE4 is to not do things like that yourself, in my opinion. I could write my own little engine otherwise.
Physics multithreading is essentially included in any engine by default, because it is a very cpu stressing application.
Yes, PhysX itself is multi-threaded, it just seems the UE4 task scheduler does not distribute very well.
Is Ori Cohen referring to the async scene? When I run physics in UE4 on my Intel i5/Win8 and the simulation gets heavy, then the frame rate goes down to slide show level and there is no sign of using more cores. So usually 2-3 Cores of my quadcore are not used. (no async included, which is destructibles limited I believe) and I look at a slide show.