My understanding of the difference between mastering at 48 versus 44.1 has more to do with the analog anti-alaising filters in front of the A/D stage than it actually does with the digital difference.
Theoretically, 44.1 can reproduce frequencies up to 22.05khz and 48 up to 24khz. However, since there is no such thing as a "perfect" analog brink wall filter, it has to start rolling off early, to make sure that the analog signal is completely attentuated at the Nylquist frequency, to prevent alaising.
So, the filter in an A/D set at 44.1 starts to act on the analog input at 20khz, and rolls off so it can completely attenuate the signal by the time it hits 22.05khz. When the A/D is set to 48khz, the filter starts to act on the analog input signal at 22khz, and rolls off so it can completely attenuate the signal by the time it hits 24khz. You will notice this where recorders show frequency response specs for 44.1 maxed at 20khz, and 48 maxed at 22khz, to account for the analog filters impact on the signal.
Looking at even higher sample rates, the filters are even less taxed. This logic to me leads that a signal recorded at 48khz or higher when resampled in a computer to 44.1khz will have none of the roll off that a signal recorded at 44.1khz could have, depending on the quality of the anti-alaising filters. This means that the resampled signal to 44.1khz can attain the theoretical linear 22.05khz maximum, where the mastered 44.1khz signal's response will be flat only up to 20khz.
In practice, I have tried running 44.1khz for the 1st set, and 48khz for the 2nd set, and resampled, the 48khz sample has cleaner high end that seems more extended than the 44.1khz master. I think if you a using software with a good high quality sample rate converter, using 48khz or higher is best, to my ears. And, my ears being human, it is physically impossible for me to hear any signal above 20 khz anyway, so amost all of this discussion is mostly academic.
Hope this helps!