I've spent odd moments wondering how this two-motor setup could be made to home in LinuxCNC (which doesn't support it out of the box). What I think should happen is:
1. Both motors head for home at "find home" speed.
2. One home switch trips. Stop both motors; drive both motors back at "slow home" speed until same switch resets. Now one side of gantry is correct.
3. Second motor is sent towards home position at slow speed until its switch trips. Now drive it back at slow speed until its switch resets. Now both sides of gantry are correct.
In principle, this all sounds OK, except for two points. First is what happens if the gantry is really, really, stiff - this homing process is trying to deliberately rack it. Second is what happens if both switches trip closely enough together that the software can't distinguish it. You could choose one motor arbitrarily and then move the other one away from home a very small amount and then it's back to step 3. Or do you say, "Close enough - let's call this homed". Don't know what the Mach3 homing routines do.
I think that the answer to your question 2 is yes, you do have to rely on switch repeatability. However, Jazz did some measurements a little while ago (in response to some comment of mine) and they do appear to be surprisingly repeatable. I seem to remember 10 microns or so, which should be good enough. Answer to question 1 is that however you home it, assuming that it wasn't spot on first time, and then rehome, you are going to hit this situation - so my hypothetical "both switches trip together" situation is almost certain to occur fairly often. And that's the answer to question 3 as well.
I'm not even sure that this two-motor homing is possible in LinuxCNC due to the way that the internal motion control logic works with two motors - don't know if you can execute the homing logic outlined above - so it's all a bit theoretical for me at present. Guess who's going to go with single motor/belt-drive to two ballscrews, at least in the first instance...