I find it hard to believe that will achieve a permanent accuracy of 0.03 C with a PT-100. A standard class A type has an accuracy of roughly 0.15 C; the most exotic type I've ever seen was a PT100 from Isotech with an accuracy of 0.001 C, but that was the sensor alone! Imagine the hardware you'll have to built around a sensor like that to get even close to the accuracy you would want to obtain, let alone the price ....
With respect to the Dallas 18B20: we are currently testing some of these to see what the long-term stability and repeatability is. The idea behind it is to produce a correction for a sensor so as to increase the accuracy (just by means of a simple polynomial proximation formula where the parameters of the function are supplied with each individual sensor). If you are interested I can inform you on that as soon as we have some reliable test results.