It's an interesting subject! Let's bear in mind that the velocity of the hammer is determined by the velocity of the key, and in the few mm of travel between the hammer escaping and the key hitting the key bed there isn't anything that can affect the velocity of the key significantly. What must be important, then, is the momentum behind the key, which is in turn means mass of finger/hand/arm coupled to it. (Momentum is (mass) times (velocity) and we've just said that velocity is fixed so we've looking for a variation in mass. Obviously momentum is important - this is basically just saying that a 1kg weight hitting the keybed at a certain speed would make more noise than a 10g weight at the same speed.)
Depending on how stiff the muscles are as the key goes to the bed, the coupled mass can indeed vary. At this point, though, it all gets terribly complicated because there are lots of places where the muscles can decouple, and it's not 'on' or 'off', it's an infinitely variable scale of coupling. All the same, one would think that the case where the finger is at least somewhat decoupled before the key hits the bed would give less 'spurious impact noise' (for want of a better term) than the one where the arm is stiff. And certainly, this will be at the start of the note. But how important is it really?
I think that can only really be answered by experiments. Here's one I propose to try out. I will pick a note at random (A440 would be suitable, I think) and play it lots of times with two different techniques - first, with as 'soft' an approach as I can manage, and second, with my arm as rigid as I can make it. I'll record this (I've got very good professional recording equipment in my music room) and I'll select pairs of impacts (one 'soft', one 'hard') that are as closely matched in level as possible. I'll encode them as high quality MP3 and post them on here.
Does anyone have any suggestions that could improve the experiment? Nothing TOO time-consuming, please!