Are they linear or exponential?

ie -you have a RF with a 75 arm strength. You upgrade to a RF with an 85 arm. Then up grade to another one with a 95 arm. Is the 10pt increase the same, both times. Or is going from 85 to 95 have a higher return on investment.

No one has answered this because no one can answer this.   But there has to be a point of diminishing returns.

The best way I can think to explain it is as defensive plus plays.   If you put a true 1B(40 across the board) at first, he might make 5 + plays.   A LF(60 across the board) might make 15.   A 3B(70s) might make 25.   A 2B(80s) might make 30.   A 90s SS isn't going to make 40(at least I haven't seen a 40 yet).   

I think it's more important to think of ratings in terms of how they compare to other players. In the instance of arm strength you give as an example, a 75 Arm RF would be league average or slightly above. 85 would be excellent. 95 is an arm you'd likely see only on a SS.

The other consideration is that very few ratings are definitive by themselves, they all seem to affect player performance in combination with other related ratings. Part of this game it seems to me, is learning or guessing which ratings work together.

And its nearly impossible to measure the results. Even though they count assists they don't tell you how often the runner successfully took the extra base on a close play where the RF's arm was a factor. Nor do they tell you when a runner doesn't attempt the extra base cause of the RF's arm ratings.

I've never tested a team of strong armed OFs, doesn't really sound particularly interesting to me.

Now putting a 90 range ss at 1B in LoA ball to see how many + plays he can accumulate,  that will have me getting up early to check the box scores!