Irmin Not sure I understand exactly what the gain response curve is doing.
It applies different (than the recorded) volume for the velocity layers. To understand it, let’s assume for a moment that they recorded 127 separate velocity layers. Each one has a different timbre and different volume. You can assign one layer to each MIDI velocity.
The curve on the left will reassign those layers to different MIDI values depending on the curve.
However, if you think that you like the timbre change but don’t like the volume change, you can change it with the curve on the right. If you make it a horizontal line, this will effectively make it a completely compressed piano sound where ppp and fff will sound with the same volume but preserve the timbre change from soft to bright.
A huge grand piano has huge dynamic response because of a big and heavy action. However, a light DP action when controlling these samples may lead to unnaturally jumpy volumes since the light DP hammers can’t be used in that grand, they wouldn’t reach the highest velocities. In this case you can compress slightly the dynamic range of the samples but preserve the timbre variation. Please note that this type of per-sample compression is more effective and natural than audio compression since the latter will dynamically affect the entire piano sound, altering everything simultaneously whereas the per-sample gain applies constant correction only to the specific sample.