Adjustable FoV is a must, but there
is a performance caveat that I'll come to. Adjustable seating position may help, too.
The problem with a single screen is that the FoV you choose will be a compromise between peripheral vision, timing cues etc. (high FoV) and the high-res feedback of the car's attitude and distant cars and other cues down the track (low FoV). That compromise is different for different people, probably in different cars on different tracks.
Many people can't afford three screens, in terms of cash or space, but they really do make a difference. I used to have three screens for PC sims a few years back, but don't have the space now, and couldn't justify the expense of three PS4s anyway... I miss the comparative lack of compromise between conflicting ends of the FoV scale, but I'm hoping Head Mounted Displays can help there (although it's potentially a different kind of compromise). Cockpit views are laregly useless to me until I can surmount said FoV compromise.
On to the performance issue:
But at the same time, everything will be smaller, and require fewer polygons and lower res textures in order to look as sharp and smooth as with a narrower field of view.
In the best case, the number of polygons added to the render list will be quadratic in the FoV angle, possibly cubic for areas with a lot of "verticality". The resolution decrease is only linear. So increasing the FoV generally requires more polygons to be drawn, unless you want a
noticeable reduction in visual detail. Luckily, usable FoV settings only cover a small range.
...
You might see more stuff on the screen at once, but it should be rendered anyway. You could look left and right by moving the camera about or looking behind you, so the game's got to have everything prepared you'd get to see by doing so, regardless of your FoV. The performance impact should be minimal, if there actually is any.
Changing the FoV doesn't seem to have any impact on the PC games I did it with, but I'm not 100% sure whether that's because it's expected the user might do that. Either way, it should certainly be possible without having too much of an impact on the game, its performance or the development cycle, I'd assume.
The game keeps a list of the "currently visible" polygons and only renders those, regardless of how many are in memory. In the most basic sense, anything outside the view frustum is culled from that list every frame; maintaining that list is a cost that is just dealt with and the performance we have is in spite of it (or, rather, because its benefits outweigh its direct cost).
You'd only see an impact if the game is polygon count limited, incidentally. Which PC games typically aren't (usually pixel limited), but any GT on PS3 definitely is. Some games also account for FoV via a global LoD bias: higher FoV, set LoD swapping to lower detail items sooner. In GT, a lot of the screen effects are dependent on the FoV, and presumably making these dynamically take that into account requires too much extra per-pixel processing.
With a continuous LoD system, such as the new adaptive meshes and "tessellation" allow, especially should they be applied to scenery as well, the poly count can be more accurately controlled and scenes can be more versatile and dynamic as a result: including, potentially, FoV changes.