If models that are 100% complete in every detail exist at all, they'd be so large as to be quite unusable in a large scene.That's if they exist. Generally I'd imagine that detailed models fall into 2 categories - those made by professionals with a particular purpose in mind (advertising/marketing, CAD/CAM design and manufacture etc) and those made by hobbyists just for the sake of making a cool very detailed model. Both groups tend to focus on what interests them and leave out stuff that doesn't. The hobbyist who models the inside of a gearbox including the the little spring under the lip of the oil seal (because he has one open on his workbench for reference) isn't going to bother to make a car around it at the same level of detail because life is too short. The car manufacturer who produces detailed models of every part isn't going to bother to model every crimped-on terminal on the wiring harness because they simply don't need to see those. Both types of model will have too much detail in some areas and not enough in others.

Both types are made with a particular focused intention, localised detail for its own sake for the hobbyist, particular selective details for the pro, with no consideration given to optimizing the model so that it can be used as part of a large, complex scene (such as a game asset would be made).

Something like the car would probably have been modelled using a NURBS or parametric solids modeller (like Rhino, ProEngineer, Solidworks, Inventor etc) and even in its native format it would be a monster file with a million layers and a million material definitions. Converting that to a polygon model would be just the start of your problems - the process is never straightforward, successful conversion between formats generally requires fluency in both applications and a full understanding of the strengths and weaknesses of each. Almost every nurbs part would require different export tesselation settings to achieve the right balance of enough polygons to define the shape but not so many as to overwhelm the system. Modelled detail like threads on bolts and studs would probably overwhelm anyway, so you'd have to re-work all that as bump or displacement maps in your poly model. The original application probably made extensive use of instancing, clones and proxies for repeated elements like screws and bolts to keep the file size down, you'd lose that optimization during conversion.

How would you see inside it? By means of physical cut-away booleans, or render booleans, or peeling away geometry, or changing opacity? Each of those would require a different approach and the way the model is set up would be quite different for each.

I think models for a project like this need to be made specifically for the project. Optimization is everything, keeping one eye on the bigger picture all the time, adding detail only where you need it. You simply don't need 100% detail on every part because you won't see it all, there's no way to see every detail of every part. You only need enough detail for it to look right in the scene, and that's a lot less detail than you think.