Do them all in series and you save a lot of I/O but it means also that if something trips it's not easy to identify what exactly did it (in theory) because any one or any combination could be tripped, or stuck, and you can't tell which. When the machine homes then it will have to do it one axis at a time. In parallel then you use more I/O but the system can see which limit tripped exactly and it can home all axises at the same time. In reality it's probably going to be rather apparent which one tripped it and the sequential home process is not THAT long anyway.
With EMC and soft limits there is little real need for a positive limit sw since the machine will check the program for any exceeded travel and the motion planner checks again before each move. The only way to run into a limit is to run without homing or get into trouble and stall the stepper motor and then it gets out of sync enough to run into the edges even though it thinks it still inside the limits. Even if you do, steppers tend to stall out and won't often break things on the machine unless you gear them down for slow speed but gobs of torque or rapiding around at warp speed.
I imagine Mach3 is similar.
CNC: Making incorrect parts and breaking stuff, faster and with greater precision.