That's odd - the software calibration routine should measure and compensate for any runout (even if the ruby tip is slightly out-of-round). The measured values are stored within macro variables #558 and #559 (for X,Y respectively).
Maybe this is a dumb question, are you using a good round gauge ring when you run the XY calibration? Something has to be changing, whether it's the machine, probe, or anything in between...
I will say this though, the probe's accuracy depends on stacking up a lot of variables, so a good mechanical dial indicator is still the last word in accuracy. If I absolutely had to measure a feature below 0.001", I would probe it first then check it with an indicator to verify, because there's no visual way to know whether the probe's reading is *that* good. The only way to check it is to test with an indicator or run the calibration...but after you do that, it defeats the purpose of the probe being reliable for every scenario. So the probes only go so far... the extent might stir up some debate