What Ed said. You really do want the greater bit depth in order to maximize resolution and headroom all the way through your signal processing chain. Then, you dither down as the last step before burning.
The quality advantage you get from 24bit is *much* greater than the quality advantage you get form the 48kHz sampling rate, given modern oversampling converters. However, it's all moot if you have a bad dither algorithm (or worse yet, if your software simply truncates down to 16 bits without dithering at all).
Absolutely do try 24/44.1, us a good dither algorithm, and see how you like the results. Straight dither is easier to do than sample rate conversion, and not all software packages get either one right...