I think of the riving knife doing two different jobs:
1. As
@TomCT2 puts it, "... prevent the kerf from collapsing on the blade ..." Sure, there may be some twisting action, but I like to think of this as the freshly cut wood "pinching" together, like tweezers.
The riving knife/splitter must be thinner than the kerf, or the cut wood would bump into it during normal cutting. Here is the rub with that (pun intended): If the riving knife is thinner than the kerf, then why doesn't the blade pinch the blade anyway? My response is: It does, but the riving knife limits how much "pinch" there is, so that you can retain control of the workpiece if it pinches.
2. The other thing that a riving knife does is prevent you from turning more wood into the back of the blade, especially after the workpiece is cut apart. Imagine a push block on the right side of the blade, after the workpiece has cleared the front of the blade and is cut apart. Without a riving knife, you could easily rotate the workpiece counter-clockwise (anti-clockwise) on the table and give the back of the blade a good "bite" of wood to trigger a kickback. The riving knife help to prevent that, too. Again, the longer the riving knife, the angle is shallower, giving less wood for the back of the blade to bite. (Yeah, sure. Everybody pushes perfectly straight, keeping the wood directly against the fence at all times. Nobody sneezes while cutting.)
My belief is: The longer the riving knife, the better. Obviously you are limited by what will fit in the insert plate hole, and other mechanical limitations.