Your specs are pretty ambiguous.I need to vary the voltage indipendently. The voltage doesn't need to be too constant (a few mV up or down). It will be small currents, less than 1mA. It is about 8 of those circuits, voltage range will probably be 18V, value of R1 in tens of Kohms, required accuracy is about +/-20mV, overvoltage will vary, but max 9V. Temperature and current dependency is neglible. TL431 gives me weird results in spice, as it apparently needs to stabilise, resulting in a spike on the waveform, thus unusable for AC, I'm guessing. The most important thing is the ratio between voltages that multiple circuits will be at.
When you say "voltage range will probably be 18 V", the voltage range of WHAT will be 18 V? Vin? Vref?
And what will that range be relative to? Vref? V-? The midpoint of V+ and V-?
You say that the voltage doesn't need to be too constant. Which voltage?
Then you say that it can only be a few mV up or down yet later you talk about a required accuracy of 20 mV.
What is this overvoltage that you talk about?
You simulations in SPICE might not be giving you stable results because you might not be letting the start up transients die away before making your measurements.
What is this "most important" metric that you are talking about? You say it is the ratio between the voltages that multiple circuits will be at. That seems to make no sense given that you want to control Vref independently in each of the circuits. So even if you applied the same Vin to all of the inputs, the outputs will depend on what the Vref is for each circuit. What, exactly, is this ratio that you are talking about (since it's the most important thing, it seems like it should be clearly defined, doesn't it) and what is the spec for what is acceptable and what is not acceptable in this ratio?