I tend to disagree with rdrast on this point.
Most of the winders we manufacture are velocity controlled center winders. This is for two reasons. the first is that the velocity loop provides a significant amount of damping to the mechanical system (motor/transmission/roll) that isn't there in torque mode winders. This allows the actuator system to be relatively stiff and provide pretty decent disturbance rejection. Secondly, with low tension systems your tension torque is often lost in the system friction torque when the roll diameter nears core. At that point you tend to lose your ability to stay on tension because the torque required to overcome system losses is large relative to the torque required to maintain tension.
I won't deny that implementing a velocity center winder is more difficult than a torque based center winder, at least in terms of maintaining stability. But if you take the time to make sure the baseline spindle speed command is correct it goes a long way toward making your life easier.
One of the big issues with load cell based tension systems operating at the edges of the tension range is that the error magnitude is not symmetric. Tension cannot be negative. So if you have a 0.1kg tension setpoint you can only generate 0.1kg of "loose" tension error. Conversely, given your setup, you could generate up to 9.9kg of "tight" tension error. This requires a relatively high gain to allow for corrections when the web is loose but can generate serious corrections when the web gets tight. This can really throw a complication into your integral sum. This is why you may need a gain change at low tensions. I know this is a no-no in continuous control circles, but on some systems I have intentionally limited the magnitude of the error to be no more than the least available signed error. So in the case of your 0.1kg tension setpoint I limit the possible error to +/- 0.1kg. This allows a more symmetric bipolar response.
Another reason you may need to change gains at different tensions is based on why you are changing tensions in the first place. Generally speaking tension setpoints are reduced wither because the material simply can't handle the tensile stress or because the material is so extensible that the higher tensions cause wind quality issues due to the compressive stress created by the linear strain put into the web as it is wound. I will assume the second is the reason you are running lower tensions. If this is the case the webs run at lower tension require a larger linear displacement to produce the same change in tension compared to a less extensible web. This is, in effect, a change on system gain that you need to compensate for with your controller gain to make the system react as it did before.
So, long story short, changing the gain based on tension setpoint is not that uncommon for a couple of reasons. If it works for you, I like option 3. That concentrates the change in the area you know is an issue without affecting the areas that aren't. While the gain may not be optimal in the 0.7-10kg area if performance is acceptable I wouldn't mess with it. You may be able to even make a few discrete gain changes in the 0.1 - 0.7kg range and not even worry about linearising the gain curve.
Keith