I almost always record at 24/44. Why? Because I know I'm going to be listening at 44k most of the time (CD's or 16/44 flacs on my iRiver). I think about the mathematical process of resampling from 48k or 96k down to 44k... all that mathematical interpolation to get to 44k, I think it has to sound more accurate if I go direct at 24/44 than if I record at 48 or 96k and resample. My "proof is in the pudding" test was taking a 1bit recording from my MR1, letting audiogate save it to 24/44, then again at 24/48. Resample the 24/48 to 24/44, invert and combine with the 24/44, so all that's left is the inaccurate little crumbs left over from resampling. There is a lot more left over than you might expect.
Back when I had Earthworks (which are documented to be able to pick up soundwaves up to 30k), if I was recording onstage I felt good about recording at 24/96 thinking I might actually pick up some HF information above 20khz. But if I am recording a PA, I don't believe there is anything above 20khz hitting my mics that I might capture, except possibly the gentle harmonics from "loud whistling guy" and "loud clapping guy". They are recording the show with SM57's and CM81's, running through the board, amp, and PA... no way any HF information is making it to me.
Jon has a good argument about the quality of brickwall filters... which is better? the one in my rig, or the one on my computer? I don't know.
And I can't really hear 16k any more. Not so much hear it as "feel that there is something there".