There are other important reasons for higher sampling rates than frequency response relating to Nyquist–Shannon. I was pleased to see someone earlier mention higher sampling rates and apparently thinking ahead for posterity, as I often find myself alone in thinking like an archivist in these discussions. We should always be looking beyond our current state of the art with a view to preserving the best possible recording we can for new and better ways of capturing sound.
I approach recording quite differently than most people here because I heavily integrate analog in my studio. 24/48 is my standard. There's not so much detectable difference between 20-bit and 24-bit as there is between 16-bit and 20-bit. For audio in music recording 20-bit is more than adequate in most cases, but 24-bit is there and has been standard for some time now, so why not? That's about it when it comes to bit-depth. But I was perfectly happy when 20-bit or even 18-bit was the bee’s knees.
With sampling rate I recommend 48k regardless, but that's partly because I avoid D/D conversion, preferring resampling from 48k to 44.1 through the analog interface instead. In my experience and IMO, digital conversion from one sampling rate to another up or down does more damage to digital audio than does resampling the analog signal through a decent ADC. There’s an abundance of evils in converting back and forth between sampling rates within the digital realm and I believe this is the area more than any other where digital audio deteriorates and gets it’s bad reputation.
A typical process for me would be to track to analog multitrack, transfer those tracks to digital 1-to-1 @ 24/48 and then mix to ¼” half-track reel-to-reel analog. From there I master to either an HHB CDR-850 stand-alone CD recorder or back through my DAW interface using CD Architect. Depending on the project I may skip mixing to half-track analog, but everything else remains about the same. I use few plugs… very often none. I have a wall of outboard gear so I realize my methods might not be practical for everyone. But I’ll keep doing it this way for the foreseeable future. I’ve found the difference between my process as described above and doing everything ITB is a night and day, heaven and hell difference.
So anyway, all things being equal, 24/48 is the sweet spot for me when working in the early stages of the digital process. And of course 16/44.1 for the end product because that’s what we’ve got at this point, though things are changing (for the worse unfortunately) rather quickly. Even so I find converting from 16/44.1 to mp3 is a bit better fidelity if we can use the term fidelity when talking about these lossy compressed formats. Others are certain it’s better to convert to mp3 from higher bit-depth and sampling rates and that’s fine for them, but I’ll keep doing it this way until something comes along that really shakes things up technology wise. Perhaps if I were tracking initially to digital rather than analog I could get more excited about 96k, but at present it gives me no advantage.