A different value for R2 and R3 is definitely an option. The data-sheet already has circuits with 30 K and 50 K as examples. There is also no real need to have the same value and type for R2 and R3 - the main reason is a simpler BOM. More current at the transistors gives slightly less temperature effect from the transistors (which is not good), but with only 50 K the differente is minute ( some 1% for the temperature sens and some 3 ppm/K for the unheated TC). On the upside the noise is marginally lower. A somewhat higher values (e.g. 100 K) would work as well, but chances are that there is not much difference. With a much different value (e.g. 300 K or 20 K ) one may have to adjust some of the capacitors too.
5 ppm/C, 10 ppm/C and maybe even 25 ppm/C would still be acceptable for the resistors. The relevant question is more the long term stability and a low TC is used as a overall parameter to judge resistor quality, as specs for the long term drift are rare and vague. The drift is attenuated by a factor of some 100 to 500. So 10 ppm/C for the resistor would lead to some 0.02 to 0.1 ppm/C for the voltage and thus likely only a small contribution. If the same resistor series is available in 2, 5 and 10 ppm/K grades it may not make a big difference to use the best selection there.