In our game we are stacking a number of object on each other and it is possible to hit them with a 'bullet'. Normally if you hit the lowest object, the stack will tumble and all objects will fall on the ground. That is, if you aim properly

Under certain circumstances however, it is possible to hit an object in such a way that objects that are stacked on top won't be affected by the impact. Effectively this means that it is possible to remove an object from the stack, while the others that are stacked on top of it stay untouched and in the air. We had several occasions of these magical floating objects. We also noticed constructions that aren't physically possible because one of the supporting beams was shot away, yet the structure 'magically' still stood there because the structure didn't 'wake up' from the impact.
With a bit of hackery I managed to cycle through all inactive objects, 'touching' one inactive object per frame by activating it for one frame. With 60 inactive objects that would mean that each object is at least awake once every second. If the collision response produced a situation as I described above (a floating object) at least that floating object is active once a second. And as soon as that object 'discovers' that it is floating in the air, it'll fall. It is hackery in its purest form, but hey... it worked...

However a major drawback from that approach is that the physics world is constantly active, and is also actively colliding. When activating an object in a group (island) ALL objects from that island are suddenly activated, which is a performance hit. But even worse, when we played a sound at a certain impulse threshold, we noticed that with the cycle approach some levels were constantly 'clicking'. Apparently when briefly activating the objects, they can collide and generate quite big impulses. These big impulses are still generated if the objects are stacked on each other for minutes: they never seem to really settle, as soon as you activate them they collide again.
That's when we decided to abandon the cycle and activate approach, we simply got too many bogus impulse results. But now we got our floating objects back

Help!

Cheers,
Martijn