I’m detecting a whole lot of tension in this thread, and I think perhaps some persons here may have temporarily forgotten the whole point of why we’re here. Lets remind ourselves. We want to tell stories with pictures, compelling pictures that move in convincing ways. To that end, we come here to discuss, to exchange knowledge and technique. We want to be competitive, but we must remember that this competition is among friends. There is no actualy monetary prize here. And everyone here, even the newbies, has something to offer. There is little point to withholding information about a particular technique, as the person from whom you are withholding probably knows something you would find useful (or will at some point), and you only make it less likely that they will assist you in the future if you withhold information now simply for the sake of winning what was supposed to be a friendly competition. It is similarly pointless to hurling insults or invectives when someone does choose to withhold such information. So lets all have ourselves a nice big group hug, and move on.
Now, lets look at ivanisavich 's animation to see why we find it so compelling. The first thing I notice is that the camera work is very natural, it appears to be handheld from on board another vehicle moving along side the crashing car. When the crash occurs, there is a sudden jerking motion, as if the camera man is in shock and attempting to track the rapid movement. We get the sense of a real point of view, a real place in the space of the scene. The rendering is also very accurate, with soft shadows, light bloom, motion blur, etc. And the sound gives a real visceral sense of impact to the image that we feel as well as hear and see.
The next thing we notice is that the overall motion of the car looks very realistic. ivanisavich said it was straight reactor rigid-body dynamics, so if you don’t know these very well, read the built in tutorials, and look up others. Further, try to use real-world values for size, mass etc to get the most accurate motion possible. Reactor allows for connections that break when you instruct them to, or upon a threshhold force, so you can give your car as many break-off pieces as you want, resulting in the spray of bits and pieces that are so convincing here. Or you could even use instanced particles for this (It’s hard to tell in ivanisavich’s animation where the individual pieces come from on the car because the impact occurs rapidly- in such a situation, particles might be easier.) In general, the meshes used for rigid body collision should be simpler than your rendered meshes (to speed up calculation) and slightly smaller, to give you room for deformation (see below).
Only after we see these things do we notice the deformation that everyone here is discussing. This deformation occurs relatively quickly, over only a few frames. It could easily be done with a few blend shapes, FFD’s or other deformers. The cloth system can also give nice results, but it’s not necessary. If you choose to use it, I think the basic technique is to bake the rigid body simulation then pin the cloth to the rigid body model every where except where it needs to deform. The mesh used for cloth deformation should also be lower res than your render model, but should match the topology, not being larger or smaller. You use the simulated cloth mesh to deform the render mesh. You only need to simulate the cloth over the few frames right during and after impact- when the physical strain is greatest. For the rest of the time, the whole body should be rigid. If you have separate pieces coming off and hitting the ground and bending then, you could also use cloth solving for the few frames around each of those impacts. But you might be overdoing it.