The testing devices at police stations are calibrated for temperature, which does make them far more accurate.
They still make estimates based on partition ratios, which does introduce a very real error. Granted, for the average DUI case, if someone is measured at 0.12, it hardly matters. If the error caused someone that was in fact .01 to measure .021, when their job is on the line, I am not sure it is really a reliable test to use.
I drink a lot of orange juice, it is possible to blow in the .01-.02 range from that.