Two fold. One, the filter behavior improves, hopefully pushing any aberrations out of the audible spectrum. Second, the DAC chip itself may perform better with more data, even if that data doesn't actuall increase the content of information from the original.
Upsampling is not all the same however, there's linear extrapolation and curve fitting with the latter requiring quite a bit more math and compute horsepower to do right.
In the end, only your ears matter though. :)