No, I think it's two different problems, with lag encouraging one of them to appear. First is body intersection, where LFS seems to want to separate the bodies by applying a force just high enough so the intersection is gone in the next collision detection cycle (which runs at 100Hz, I believe). This happens mainly if cars lag into each other, shooting both off into the opposite direction. Some objects seem to have the same basic reaction, but the force is obviously dampened to make them soft (barriers, tyre stacks).
The second, very specific problem is when the contact patch of one of the tyres starts intersecting with an object. I think LFS always tries to put the tyre on top whatever it intersects with, which is okay for bumps on the road, but has quite spectacular results if the "bump" happens to be a red-white barrier. The force calculated from a height difference of a metre within one frame does the rest.
Example 1: South City sideways slide into wall. The tyres on the impact side start intersecting briefly with the wall, generating a short force spike upwards on that side of the car, which is why it always results in a roll "away" from the wall.
Example 2: Red white barriers. These suffer greatly from both problems, with the main quirk being that they are "soft", but at the same time unmovable. Soft alone usually means that with enough force you can drive through them, because the counter-force is not 100% of what is needed to separate the objects, which in the case of tyre stacks is barely noticeable as they're shot off themselves. The barriers however stay put, making it much easier for you to overcome this repellent force allowing you to pass through them with relative ease. For "shortcut" purposes it is generally enough if you have enough speed to make the car pass through the barrier halfway, because as soon as you're through more than 50% it starts shooting you off into the direction you actually want to go.
But with passing through objects, there also comes the risk of your tyres intersecting with it during a collision detection cycle, which is the reason you can only pass "safely" through them with a certain speed, and also explains why the reactions of passing through are seemingly random. Just as an example, imagine we barge through a barrier with 100km/h, which is 27.7m/s, which at 100Hz means that the collision detection makes a check every 27.7cm of car movement. Now picture a side view from the car, drawing a line every 27.7cm. If one of these lines intersects with one tyre's contact patch, you'll fly off, or if it's barely touching, suffer from suspension damage. Of course you cannot calculate a perfect passing through speed with this for each car, because the first line is not necessarily drawn at the front of the car, but can be placed randomly within the first 27.7cm of it, depending on how long ago the last collision check ran when you started intersecting the barrier.
Easy to analyse? Yes. Easy to solve? No.