48kHz vs 96kHz: audible?

I have found similar situations but with streaming from Qobuz. I regularly go in search of new music and pay no attention to the sample rate when selecting songs. Many times I’ll be listening and the difference in how smooth the tones are makes me check the sample rate and it’s almost always 96 or above. However, if I then start to try and replicate this actively I find sometimes I’m assuming a CD quality file might be high res. There is a definite difference for me when I’m just normally listening, but it clearly depends how good the mastering is of a CD quality file as some of them can sound superb. Whenever my ears perk up to check a sample rate it’s always because of how much smoother it is, I’ve always just assumed more data means a smoother crescendo of sound, if that makes sense.

It is possible that JRiver Media Center's downsampling was done poorly and became more noticeable than if it had been done correctly. A lot of what we "hear" comparing different resolutions is dependent on how familiar we are with the music being played. Obviously you know the Holst very well, but would you have noticed the difference (or any difference) on a less familiar one?

This whole area is a bit of a minefield. The ideal way to test this hypothesis would be to take a piece of music that can be mechanically performed -using something like the technology that was used for the Zenph reperformance of the Goldberg Variations. One would then have the computerised piano play the piece of music twice and record it at the two sampling rates/word lengths, and then double blind test it across a stratified sample of listeners using a high resolution playback system.

In other situations where upsampling and downsampling is involved, it's hard to discern cause and effect.

Qobuz is my sole front end, and I hear a slightly more refined attack on each note on most hi-res selections. However I find hi-res content overall of inconsistent quality and less the driver of SQ (compared to Redbook) than the recording/mastering put into it, so much so that I no longer even care what its published resolution is, FWIW.

Hmm. I suspect that what you were hearing may not be SQ difference between sample size, but problems in how your box downsamples digital signals. 
 

Downsampling problems show up in edge cases, like high violins, reverb tails, etc. For downsampling to sound good, dither (digital noise) needs to be used to minimize aliasing/rounding errors during the conversion. There are several ways to do this, which is best done when mastering the recording. Sounds like your streamer might just be lopping off bits on the fly?  And you heard it.  Might be interesting to find out how your box downsamples and what kind of dither is used, if any. 

A lot of problems can be introduced when sample sizes get changed without close attention.