OM,
1.) You are probably opening a can of worms by using FETs for the current booster. In particular, high frequency stability and possible damage to the FETs by exceeding their gate to source voltage limits. I'm surprised the bipolar transistors do not have a graph of the Safe Operating Area. What devices were you using?
2.) Without going into great detail, I'd say that by not buffering the + divider, you'll get a current error in absolute terms equal to Vload / 80K. This is because the divider is diverting some of the current pump's output to circuit common. If you want 100 mA at 20 V, you'll get 99.75 mA. If you want 1 mA at 20 V, you'll get 0.75 mA. If this is used in a closed loop servo, this is most likely acceptable. IOW, the non-linearity just gets lumped into the stiction of the actuator. Another way to look at it is that current pumps should have infinite output impedances, ie., the current is independant of the voltage. This will put an 80 K resistor in parallel with the current pump.
3.) if you decide to use an opamp to buffer the + divider, good design practice is to not exceed its input to supply rail voltage and current specs. It'd probably be acceptable to use a resistor of say, 10 K to 100 K in series with the + input of the buffer to limit current in case a greater voltage is applied. This will introduce another error due to an offset voltage developed across this resistor by the input bias current of the opamp. (see Final Note)
4.) If the INA is powered by +/-12 V, there won't be any real difference in the transient response. An inductance generates a counter voltage across itself equal to the rate of change of current thru it. Therefore, if you want to change the current faster, you need to apply a greater voltage to overcome counter voltage. Using +/- 40 V on the current booster does not change the fact that the INA output is still limited to +/- 12 V. The current booster has a voltage gain of almost 1 but a current gain related to the beta of the transistors.
BTW, I've been using the term INA for Instrumentation Amplifier. It's probably not completely accurate as these parts are maybe better described as Opamps with Integrated Differencing Divider Networks - OIDDN - or something like that.
Final note: As a Jewish friend said, "Oy Gevalt!" These things are like a game of Whack-a-mole.
Hope this makes sense and helps in some way (at least)
Cheers,