My bike has two useful throttle maps, one like the green curve, and one like the red curve.
The green one is great on the motorway, or in windy / bumpy conditions, for maintaining a constant speed at less than 20% throttle. But when you're "on it", the steep upslope near the full power region makes it difficult to modulate forward thrust around and out of corners. It's also less appealing around town, mainly when setting off, due to the large amounts of throttle movement required to do so.
The red one is more direct initially (too much so with the stock fuel / ignition map; lean fueling and aggressive timing due to emissions control), but it requires too much wrist movement to meter torque output near full throttle in the twisties, and when changing gears during acceleration (clutchless up, no quickshifter).
I've thought for a while that, ideally (for my spread of use), it'd be best with something more like the blue line. I honestly think manufacturers overthink this a bit.
Anyway...
My position is that the game should be 1:1 in terms of its input processing, purely from an input-output and player-in-the-loop feedback system point of view - i.e. in terms of response and feedback of information, in order that we can (re)adjust our input accurately. We play the game at or near 100% throttle, and that is where emphasis should be placed. Yes, initial throttle application is important mid-corner, but in a track context, that is effectively taken care of with the torque curve already.
The key as many have said is to understand exactly what we mean by "non-linear", because there's lots of ways to do it. I first looked into the non-linearity of throttle butterflies when I was messing with sound synthesis (throttle area is a useful parameter). From a control perspective, the important thing is the gradient at each position, i.e. how much power you gain or lose from a given throttle setting by moving the throttle a small amount. Large changes in output from small inputs naturally are harder to control, but more subtly and equally importantly, large changes in that gradient over the important range of operation can give a sense of disconnection.
The basic curve for a throttle butterfly is the first quarter of a (negative) cosine curve, so it gains gradient according to the sine curve, i.e. quickly from zero in the first half, then tapers off in the last half, approaching a straight line (constant-gradient). I hope that's clear in the diagram below.
To add to this, the distance (along the duct, with flow) between the leading and trailing edges of the throttle butterfly gives rise to a venturi that increases the effective area (in terms of actual airflow) more rapidly in the mid-range, yielding a closer to linear response much sooner after the initial opening than the geometry would imply.
Other types of throttle, namely vertical slides (prolific in motorcycle carburettors) and circular orifice slide plates have more complex trigonometric relations, but still have a fairly soft initial response, a linear mid-section and then can also soften out
even more at the top end, like an S curve.
Note that it assumes that the control actuation between the pedal and the throttle itself is linear, and I've put GTS on there just as an illustration, but the output is only the reported in-game throttle position, whatever that actually means. Either way, when playing, you can feel there is less control near full throttle, and I think this chart shows why.
this means no pops, bangs or farts on upshift 
