The heavier a person is, the more kinetic energy they have. Moreover, seatbelts don't really stop a person until flesh is sufficiently compressed against bone. From that, I'd reason that skinny people should have a higher survival rate. Intuition be damned, data is important.

It shouldn't be hard to spec and built crash test dummies with simulated fat.

Lower momentum one way or the other sounds like the best predictor.

Just 'spitballing', but if the restraints are adequate, the compression of flesh may act to slow down and spread out the force of the accident to a certain degree. Bruising may be more likely, and the increased overall force might break ribs and such, but increased weight may have some benefits at some speeds/in some scenarios.

Some possible downsides would be overwhelming the airbags capacity, and larger overall volume of space which needs to be protected from intrusion/crushing/etc. I Wonder how much this has been studied, hard rubber and steel dummies probably don't reflect conditions at the extremes very well...

That is exactly what the crumple zone in the car is for though. You only have X meters of margin before you need to be at full stop. Assuming (dangerous to assume, yes) it does its job perfectly the most important factor becomes to react, that is to start the slow down as quickly as possible so you can utilize the crumple zone to max effect. I feel like flesh compression will mostly delay the impact rather than properly slowing you down (fast enough), which would be a bad thing.

That is in part also why it is so important to have a properly tightened seat belt and why we have explosives to tighten it up further when a collision is detected.

Fit people also recover faster, which probably gives them a better survival rate over all. Fit people are probably less likely to linger with persistent injuries before perishing. Consider particularly the difficulty diabetics have with infection.

Right but we're not talking about energy of the homogenous object. The brain or heart or lungs of fat person won't be that much heavier, but the fat around it might work as cushion to spread the spike of deceleration over longer period (less G's over longer period of time basically). Then again, the fat on the other (back) side of the person might compress squishy stuff inside more so I dunno how well that can be simulated in a dummy