First, when we talk about velocity response, we need three main qualities: range, resolution and curve.
Though the affect of velocity on the sound is determined by the destination instrument, the velocity range of the controller is still an important part of the response experience. Though the output range is 1-127, I'm referring to the physical velocities that are mapped to those extremes of 1 and 127. I.e., what actual key-strike speed is considered a 1, and what actual key-strike speed is actually considered a 127. The lower extreme is an arbitrary decision by the designer, the upper extreme is limited by the scan rate.
If you think about your own physical interaction with a key, you can imagine your very softest possible note, and your very loudest possible note. If the controller takes your 20 softest values and assigns 1 to all of them, that's a loss of response. If your controller takes your 20 loudest values and assigns 127 to all of them, that's a loss of response.
An assumption that is being made in this thread is that most controllers provide 127 steps of velocity, which is most certainly NOT the case. Most keyboards are limited by slow scan rates (i.e., how frequently the keyboard scans for the two velocity switch contacts and measures the time between them). The slow scan rate causes some actual velocities quantized to a lower resolution, and many velocity values are simply unattainable.
Recently, Novation was touting scan rates of 10KHz for some of their recent keyboards. I don't know if that number solves everything, but it seems like the right approach.
Most keyboards will then sacrifice more of that already compromised resolution to give you a response curve.
I'd be happy if a keyboard let you define your truly fastest/hardest keystrike as 127 (and had some room above that, for very dynamic players), as well as setting your slowest/softest keystrike as a 1 (with room below that, as well). And, if the keyboard let you define a curve without sacrificing having full 127 step velocity. That is not, to my knowledge, even available in any commercial keyboard.
Our ability to hear differences in volume is very keen. We have an extraordinary range and resolution of hearing loudness differences. With typical velocity-to-amplitude applications, we most definitely can HEAR 127 steps of velocity. Create a MIDI track with series of 127 consecutive notes, playing quickly and steadily, incrementing the velocities by 1 per note. You will hear the notes steadily and smoothly crescendoing. If you decreased the resolution of that ramp to say 32 possible values, you would hear the 32 steps as the notes jumped to the next value.
And I've played cheaper keyboards with slow scan rates where I can feel a disconnect in the velocity response.
But if a keyboard gives you a wide range between a very very slow/soft strike, and a very very fast/hard strike, and gives you 127 steps between, even if you assign a curve, then this is sufficient. But that keyboard does not exist. With the possible exception of the aforementioned VAX77. I don't know the High Scan Rate Novation keyboards offer that kind of setting.
A keyboard that gives you 16,384 velocity values has plenty of resolution, but if it doesn't allow you to define a wide range between your slowest and fastest physical strikes, or letting you define the curve in a flexible way, this is not sufficient.
I agree that there is the random factor which shouldn't be undervalued. But also, there is value in just having a little overkill when it comes to the expressive parts of the instrument, especially given that our devices can handle the extra bandwidth it might require when we need it.
