Residuals are essentially the difference between the obtained result and that predicted by estimates and averages. In this case, I'm guessing this is relative to the average driver. As for the unit, I have no clue. It really depends on what system they used, and that's a study I'd very much like to read.
As for the results...ohhhh boy, it's a good example of what can happen when you try to quantify talent and rely on models to do so. You can never get results that make complete sense to everyone, because in a way, the quality of a driver (or any sportsman) is subjective. It depends on how you define a "good driver". If you define it purely in terms of success, you're not taking the car into account, nor the circumstances. Few would argue that Nelson Piquet was better than Jim Clark because he had more titles.
Then again, if you only look at a driver's performance against his/her teammate to ascertain how 'good' the car was and use that as an argument, you run the risk of overestimating or underestimating a car's quality because ultimately, if both drivers are close, you can't know if they're both bad or both good.
That doesn't mean I don't find these results immensely satisfying, though
