"Upsampling" is quite off-topic in this thread, but here's my comment anyway.
When a digital audio signal is resampled the new samples in the timeline are calculated. The signal doesn't stay unaltered. However, when a high quality resampler is used there is no audible difference. For instance, you can up and down sample an audio signal several times and you will still fail to hear any difference in a blind test when the resulting signal is compared with the original. Our ears are not very precise measuring instruments and in any case the faint distortion that is caused by such procedure will be masked by the much bigger distortions that are caused by the analog amplification and especially the final transducers (speakers or headphones).
However, resampling is a waste of CPU cycles and I'd recommend it only when it can fix problems that are caused by inadequate hardware.
I have no problems with playing various sample rates with my old Terratec DMX6 fire sound card. It switches sample rates automatically and perfectly fine using WASAPI Exclusive, ASIO, Kernel Streaming, Direct Sound or Wave Out (the last two on XP*). I have not experienced any delays or freezes when it switches between sample rates, e.g. from 44.1 KHz to 96 KHz and back again.
The above mentioned Terratec model is not available anymore, but perhaps a card like M-Audio Audiophile 192 (both are based on the Via Envy24 chip) would be a good digital interface for connecting an external high end DAC through SPDIF.
* The audio engine in Vista and Windows 7 is always set to a constant sample rate and it resamples when needed. ASIO, WASAPI Exclusive and Kernel Streaming can bypass it.
EDIT
Increasing the bit depth value is different from upsampling. If no DSP is applied it is fully lossless. When the signal is modified by any DSP the processing in MC is done in the 64-bit float mode and it is preferable to set the output to the highest bit depth that is supported by the hardware.