The only thing changing in the file being upsampled is the dynamic-range of the signal and the Nyquist frequency of the resulting file… The original encoded bits are unchanged and the sample-rate is increased by adding zeros to the bit-stream… Any other subjective assessments of contextual sound-quality are typically associated to system level DAC/filter elements and component/ aggregate system topologies that induce noise-related jitter.
Dynamic-range (gain structure) must be managed… (See my resonse to your post here: Kernel Bit Depth - #2 by Agoldnear )
