The others have covered the practicalities of Mach3 and closed loop, however closed loop using linear scales will struggle to deal with any excessive amounts of backlash.
.
Lets say for example you have the equivalent of 20 steps backlash between the stepper and scale. You command a move of say 30 steps, and want to try holding within +/- 5 steps. The controller commands a move of 30 steps, but due to the backlash, the table only moves 10steps, still 20 steps short, so the closed loop control commands another 20 steps and you theoretically stop at your requested place. But then the cutter pulls the table, and you've suddenly overshot the target point, at which points the controller commands a move back within range, and the process continues.
Another example is the table moves overshooting the target due to deceleration, and you end up with a continual oscillation.
.
Using the PID loop typically found within closed loop controllers, you can tune to handle each of these situations, but never really in a satisfactory way. Increase values to enable quicker corrections, you'll likely end up with a continual oscillation (which in the case of servos can be so severe you cause them to overheat in a matter of seconds), reduce the values to reduce how quickly errors are corrected, and you end up with an over-damped or very sluggish to respond system.
.
And that's before you consider the mechanical problems from excess backlash. With excessive backlash, climb milling will likely be out of the equation, and overall accuracy will be reduced.
.
Personally, I'd put the additional cost of adding scales/closed loop into removing the backlash in the first place. A on open loop machine with minimal backlash will always be better than a similar but closed loop machine with excess backlash.