Monitor calibration currently uses the gamma table libraries in LCMS. These libraries use unsigned shorts for the values in the table (IE. 16 bits/channel). This is likely introducing quantization errors (probably small ones) during the process of calculating the new gamma tables during the calibration process.
Using doubles instead of unsigned shorts for these gamma tables would reduce any quantization errors to insignificant levels and would increase the resolution of these tables so that they would work on systems with higher resolution gamma tables (IE. more than 16 bit/channel) when these become available.