It's nothing big in most cases; Previously the sample position and increment were stored as fixed-point numbers with a 16-bit fractional part; this was for example far too imprecise to pass some of the tests on the
SRC comparison page - hence these tests are currently N/A. In this particular case, a test sample was resampled from 96 to 44 kHz for about 30 seconds, and after that period the theoretical sample position already differed from the actual position by a couple of samples. When choosing the built-in resampler in the sample editor's resampling dialog, this problem could manifest by loop points no longer being placed correctly after resampling - but only for
very long samples.
So while this was not a problem most of the time, I have still decided to increase the fractional part to 32 bits, which is more than enough.