and do your processing in 48 bit space instead of 24 bit space, you keep more of the subtleties in the original data. As a final step, convert it to 24 bits for printing.
This page (
http://www.normankoren.com/digital_tonality.html ) goes into some detail, much of it quite technical, about this, but the "bottom line" of the argument is found near very end of the page with a pair of histograms - one (when the calculations were done in 24 bit space) showing the loss of several levels of data and the other (48 bit) showing a continuous range of tones. In short, by doing your processing in 48 bit space you get better results even after the final conversion to 24 bits for printing purposes or conversion to jpeg.
However, if you are beginning with 24 bit data, the author of Picture Windows Pro (which has been offering 48 bit processing for several years) says that there is little to gain by processing the data in 48-bit space, other than making the calculations take a bit longer. (
http://www.dl-c.com/discus/messages/2/321.html )
(edit typo)