Also, if the formula causes a large correction, watch the first move of the machine when you start a program. On my router it would hammer the motors at full accel just before executing the first move, not enough to lose steps, it was just quite harsh! I did have a fairly large error on my table at the time, though.
One other caveat is if you rotate the system (G68 or "Rotate Local System") and then feedhold, when you click cycle start again it will attempt to move to where the next co-ordinates would be if the system wasn't rotated. It will not lift Z first - if the bit is down it will rip through your part. To avoid this, i would feedhold, cycle stop, set next line, then I was OK to hit cycle start.
It sucks that such a powerful function is broken... maybe Mach4 will fix it?!?! I eventually squared the table up better, but it would be nice to get rid of the length issue (my table cuts slightly longer towards one end of my Y axis - slight mechanical misalignment of the gantry rails allowing the gantry to pivot a bit - need a double-screw setup!). It's almost square now, so I can get away with it without any further corrections.