Thanks for that Gregorio, will find some time to have a read...
All the various AES tests prove that adult humans cannot hear above 20kHz but regardless, 96kHz sampling frequency allows recording of frequencies up to 48kHz, more than double the highest frequency possible for a human to hear and actually higher than a dog can hear.
Have a read of this: White Paper
The paper gets a little technical but it's worth persevering. In Electronic Engineering the basic axiom is: the faster you go the lower the accuracy possible. Yes, communications equipment can run at GHz speeds but at a resolution of only a few bits. To produce smooth filters at 384kHz (for example) is not mundane. To produce a smooth filters without artefacts requires a considerable number of coefficients and the higher the sample rate, the more coefficients needed but less amount of time is available. In the case of 384kHz, four times the number of calculations are required in a quarter of the time compared to 96kHz. and 96kHz is already slightly faster than optimal for 24bit accuracy. The only solution is to cut corners and make the computations less complex so they can execute more quickly and this compromises the efficiency of the filters and ultimately therefore compromises sound quality. As corroborated by the link to the Benchmark statement in the link I supplied previously.
The marketing hype is getting to truly ridiculous proportions and not just with sample rates. 16bit already provides about 1000 times more dynamic range than just about any commercially released recording in history, so what's the point of 24bit? And 32bit is outrageous, if it were possible to actually resolve 32bits of dynamic range it would kill you instantly!