Dzarda's comment is valid. You have about 29uA's of current flowing through there at 12V supply and even less as your battery voltage drops. The Zener voltage will depend on the current flow through it. The lower of a resistance you have supplying current, the easier it will be to keep the Zener voltage relatively constant/stable.
You'll definitely want to look into Dirceu's comment and make sure the output of the comparator is rail to rail and isn't open-collector or open drain. If it is, adding a resistor as he recommended will work.
I think the main reason this will likely work poorly is because there's no hysteresis in it. Depending on the current draw of the circuit, you'll end up with some voltage drop due to the internal battery resistance. As soon as this limiting circuit cuts off the current, the terminal battery voltage will jump back up because no current is flowing. That wall cause the circuit to turn back on, which in turn will cause this limiting circuit to shut it off, endlessly turning on and off the circuit. This may work for your purpose, but the main circuit may get quite confused and you'll generate a lot of noise. If you have a comparator with hysteresis built into it, or if you add hysteresis to this circuit, that will be avoided.
Your current resistors show that you'll cutoff at about 10.11 volts: 9.1*1000k/900k = 10.11 V.
Lastly, although Ignacio is correct that Zener's aren't great voltage references, you don't need anything terribly accurate for your application.
Since you're measuring ratiometrically slow variations in the supply or the reference (for example with ambient temperature) will not have any effect on the reading. That's the beauty of ratiometric reading- the reference can vary 5% and the reading will depend only on the stability of the reference resistor.. ideally.
However, fast variations (for example, if the supply varies when something is switched or noise in the readings) will show up in your temperature reading. Since the resistance change with temperature of a typical NTC thermistor is very large it won't have much effect on the reading around the point where the thermistor resistance is close to the reference resistor resistance.
If you're looking for very stable readings over a wide range of temperature you may wish to regulate the 3V down to 1.8 or whatever with another dedicated regulator (there are other things that would work, but a regulator takes almost no area and is inexpensive) to keep the variations to a minimum and perhaps digitally low-pass filter at least the reference measurement to reduce the noise in the reading. Of course if you could power the resistor/thermistor from the internal reference it would be ideal, but often the chip designers don't make it externally available.
Best Answer
This answer was given before the OP kindly changed the requirements from +2 volts to - 2 volts therefore it's still under construction
Use the 2.048 volt reference and a potential divider after the 2.048 volt output to produce precisely 2V. Then using an inverting configuration op-amp running from negative and positive rails invert the 2V to -2V - possible op-amp choice is the AD8657
simulate this circuit – Schematic created using CircuitLab