Have you tried Empirical Audio's suggestion of doing the bit rate first to 24 (so saved as 24/44.1) then upsample to 96? Much more holographic presentation, although it does mean more work.
I haven't. I have a very hard time thinking this would make a difference, since AFAIK a DAC chip's registers will "zero pad" the missing 8 or 16 LSBs anyway (depending on whether the file has a 16 bit or 24 bit word length, and whether the DAC's internal processing is 24 or 32-bit; the Bifrost's AKM chip does 32-bit internal processing IIRC). Moreover, I can't see how zero padding the LSBs, whether in or prior to the DAC, would change the sound, since the resulting loudness values are exactly the same, like 1 and 1.00000000 (that's 1 with 8 LSBs zero padded). It's not like changing the sample rate, where interpolated loudness values are added, resulting in most of the bitstream (147,900 samples per second in the case of upsampling from 44.1 to 192) actually being new sound that wasn't present before.
So - don't see how zero padding would change sound, and even if it did, it's happening every time you listen to your DAC anyway.
Of course Steve Nugent has forgotten more about audio than I'll ever know, so I guess some humility is appropriate....