dddd

One consideration about bearing surface smoothness is that as the loading increases, the steel surfaces elastically conform to one another, essentially squashing down surface defects (i.e. high spots) toward the level of the low spots.

At the same time, flex in the balls forces them into a slightly oblong shape, which increases the interference contact force between balls in the loaded zone along perhaps one-third of the circumference of the bearing races. Contact here is between surfaces moving in opposite directions, so makes an argument for the use of flexible retainers which can reduce ball-to-cage contact forces and eliminate direct contact between adjacent balls.

I think we tend to view steel bearings as being hard and rigid, but under loading, all materials are elastic.

I think that when someone says that it's "how the bike rolls" that matters, it's not to say you can necessarily feel any difference while riding, but rather that it's the as-installed condition of adjustment that matters, after considering the QR's compression of the axle and the slight bending of the axle which (theoretically at least further shortens the axle and also rocks the bearing cones slightly off-axis, both of which could make the bearing turn tighter).

Often the well-used balls are still smooth and round, but this is hard to confirm quantitatively. Chromium balls tend to be "tougher" than the races they roll in.
Yet I find that using new Gr25 balls does occasionally solve a rough-turning axle even when the old ones still looked perfect.

Puzzling to me is how everybody's cup/cone hubs are shipped out with snug bearing adjustment that becomes VERY tight as the QR lever is applied. This applies to new hubs and the new bikes that they are spec'd on. I am certain that this greatly reduces the working life of the bearing surfaces. It seems unrealistic to expect finely-finished bearing surfaces to break in sufficiently over time so as to offset the shortening of the QR axle.