Hi Bill,
Basically, the correct way to treat backlash, and what everybody on this list would say is to try and eliminate the backlash mechanically as much as practically possible. On a router (or mill), if you have backlash, the cutting forces can move the table (head) to any direction within the backlash margins. Thus, the software backlash compensation can do nothing to help in this.
Having said this, many users benefit much from using the software backlash compensation. It all depends on the type of work you do and the forces acting. In my opinion, in most of the cases that you can't afford a ZERO backlash mechanical system, a software compensation will be very satisfactory solution. If you do a finish cut of say 0.05mm (or less), most probably, the cutting forces would be smaller than the friction in the system, not being able to move the table to unwanted directions.
If you have a constant backlash along the leadscrew and you can't eliminate it mechanically, you can define the value in Mach3 to compensate for. It is desirable that the mechanical backlash be as small as possible. I think that software backlash compensation for up to few hundredths of a mm can be of a huge benefit for some systems. You can cut very precise circles (up to 0.01mm tolerance) if you do several spring passes starting each one at a different location along the circle circumference. By probability theory it eliminates most of the possible errors caused by uneven leadscrew pitch or backlash compensation.
As to your backlash, I think it is too big and you should try to reduce to at least half of that, or preferably to the 0.002" region. You shouldn't be checking the backlash by the eye. The correct way to do this is using a dial indicator - you program a move towards the dial indicator and then backwards and the difference in the dial indicator reading is your backlash.
Daniel