Couple of points there. Personally (and without looking up switch data sheets to see if they quote switching times) I suspect that compared with typical machine speeds, any delay will be insignificant (*). That's based on gut feel rather than hard data, though. Second point is that I would be more concerned about repeatability than actual switching time. I don't really care if the response time is, say, 100uS as long as it is always 100uS, so that the machine always stops in the same place. Doing a bit of my usual back-of-the-envelope sums, I reckon that at full tilt my ballscrews will be turning at 1000rpm. Say, 17rps. 800u steps per rev, so 13600 steps/sec. That's about 75uS per step. In practice, final homing is done at, say, 10% of that, so as long as the switch responds repeatedly to the nearest 3-400uS, you should be able to stop at the same ustep. Dean's demo of homing repeatability using proximity switches a while back supports this.

(*) what probably matters more is that you always home at the same speed as I suspect that exact switching points will depend on speed of approach to target but that is something that's under our control. It only matters for homing anyway; don't really care exactly where the machine stops if it hits a limit as long as it stops before something gets broken!