The Philips should read more accurately than the others because a Pt100 sensor does not require cold junction compensation as for the thermocouple units.
I worked in the process control industry, and with a Pt100 a 0.2 °C accuracy was the norm, at ambient or near ambient temperature.
Ice+water+air are at the so called "triple point" of water, which is about 0 °C (but is pressure dependant) but in my opinion a standard Pt100 thermometer should read about 0 °C (not accounting for wire resitance from sensor to meter : in this case a 3 wire or 4 wire connection of the sensor may give a more accurate measurements).
Using boiling water is not a good method, the water bubbles are more hot than the liquid, and this result in erratic meaurement, and there also is the effect of the different temperatures of the water "currents" flowing inside the container.
In the calibration labs it is a norm to use heated-cooled metal blocks (isothermal blocks) where the sensor probes are inserterd (probes to be tested and a reference, calibrated probe).They tend to spread heat more efficently, and offer a thermal inertia reducing the oscillations in reading due to the above described sources.
Edit: I have two or three cheap Aneng meters: They measure randomly, even at ambient temperatures: errors of 5 °C are common, and the sensor probes differ widely one from the other... The cold juncion is, I believe, not calibrated: when shorting the input they should read ambient temperature, but this is not the case.
Best regards