And what, pray tell, is the "right" calculation? If you think no-one else is using "hacks", I think you need to have a word with yourself and think about the depth of your knowledge on this issue.
Of course people are using hacks, the computer power doesn't exist to do everything at the limit of current physics knowledge.
Perhaps I'm explaining badly. You can model a spring, a damper, suspension linkages, and a wheel separately. You can combine them and have a system that will describe the motion of a wheel attached to a car (mostly, I'm aware there's more to it in most cars but I'm trying to keep this short).
Or you can just make a lookup table for every position and force that the wheel might encounter.
The difference being that the first setup takes a few relatively simple parameters and produces an output. It's doing it's best to replicate reality, as far as the computing power will allow. Simplifications are made for complex stuff like say, fluid dynamics, but a simulator will always have some objective link back to reality.
The second requires everything being precalculated. If you're precalculating from larger sim, fine. If you're putting numbers in from experimental data, fine, until you get outside your experimental range. If you're just making numbers up that feel good...you're not simulating squat.
I think of it as the difference between a photograph of a mountain, a painting of a mountain, and an imaginary painting of a mountain.
-The photograph is what the developers aspire to create, a highly accurate rendition of the real object.
-The painting is what a simulator ends up being. Because of restrictions on available techniques and materials, the picture has to be produced by a painter with paint and canvas, which isn't ideal. It's a representation of the real mountain, but filtered through the artist's perception and limited by the way he can paint it. Accuracy may vary.
-The imaginary painting may appear to be just as real, but it's not based on a real mountain. It uses the artist's experience with mountains to produce something that appears to be a representation of a mountain, but because it's not related to any real mountain it may include features that couldn't be possible, but enhance the picture artistically.
There's a difference in how one approaches the design. One method holds real world data as an absolute, the simulation must use that data in the form it's given to produce the final result, no matter what transforms need to take place on the way.
The other has no absolutes and will use any method available to get the correct result. Which would be fine, if games weren't so complex that it's essentially impossible to exhaustively test every possible result of the physics system. That's why something that models the behaviour from a small set of variables is preferable to attempting to describe every possible output.
I'm a chemist. I create models of molecular and physical interactions. I could define every possible interaction by simply testing them all. I'm not given the time to do that, it would be hugely expensive for the knowledge that would be gained.
Instead I define a small set of experiments that elucidate the mechanisms behind the reactions and the constants that govern them. Then I create my model, and create more experiments to double check that the model is in fact behaving in accordance with the reality. The amount of work to do this is orders of magnitude less than testing every interaction, even to a very low degree of accuracy.
What's more, a good model should show edge cases and odd interactions that might have otherwise gone unidentified.
This is pretty standard scientific procedure. I don't see why it would be preferable when creating a model of a physical system to go and test everything, rather than creating a model of the system and allowing that to generate your data for you.
Maybe they did, and just omitted the final step of double checking their model against reality again. Which is a pretty big whoops.
It's funny, because GT is designed forward from first principles, according to Akihiko Tan.
When words and actions disagree, believe actions.
A system designed from first principles does not omit camber effects.
I used to work in the video business and our software was outdated. I found a company that had a ready made proven software package that met all of our needs for about $50k. The owner decided he was going to develop his own software so we spent $250k over the next two years upgrading the piece of crap software we already had in place and after two years...we had an underperforming but updated piece of crap. By that time I was out on my own and had purchased the very same software I recommended to my previous employer for my own stores and it was like night and day. 
