The engine of the PhysX simulation cant increase rpm if your vehicle has some form of resistance to it's movement. Resistance means high vehicle mass, an incline/hill, or a physical object in it's path.
Mostly affected by this are tanks because of their high mass/movement resistance, but also wheeled APC's. It severely limits offroad capability. Cars are only affected if there is an object blocking them because of their high power/weight ratio compared to tanks.
The reason for this is, because the clutch simulated in vehicle PhysX seems to be hard-tied to the controlls for throttle. As soon as you apply any form of throttle the clutch seems to be closed. Therefore, the engine can't rev up to produce torque.
The only torque the vehicle can use for starting it's movement is therefore the torque at engine idle rpm, which is close to 0.
To solve this issue the closing of the clutch has to be delayed by some time after the inital application of throttle. A delay of 0.25 - 1.0s should be sufficient. Ideally (to simulate use of real clutch) the clutch should close lineary from 0 to 1 over a time period of x (0.5-1s might be a good start).
Edit:
I guess i was wrong to say the operation of the clutch is properly simulated with physx. I did more reasearch on this topic and started a topic on the nvidia physx forum.
It seems to me as if the characteristics of the physx clutch system are responsible for this problem. In case you stand still and give thrust the resistance produced by the clutch can be so high that any acceleration of the engine is prevented. Its current function is, that it is closed when wheel-rotationspeed = 0, and opens once wheel-rotationspeed (gearratio factored in) approaches engine-rotationspeed. Basically the inverse of what you want: Half-opened when wheel-rotationspeed is 0 and fully closed once wheel-rotationspeed is increased. Here is the topic https://devtalk.nvidia.com/default/topic/764824/physx-and-physics-modeling/clutch-strength-question/ [^]
(my posts are hidden unfortunately, here is a picture [4mb] with all the text http://abload.de/img/compiledeqsy5.jpg)
Since i have not received any further answer from Nvidia i think they just ignored it. Maybe it would help if BIS Devs contact Nvidia directly rather then some random dude complaining about unrealistic Simulation on their badly moderated forum.