This all assumes you're opening a 24-bit file, editing some portion of it, then wanting to save as 24-bit once again. If you're only doing a couple minor edits - like fades - you have a couple options:
<1> Untick the dither checkbox when saving as 24-bit. This means you'll truncate the edited portions of the waveform from 32bfp to 24-bit, but if you're only doing fades or some such, you'll likely never hear the difference. And unticking the dither checkbox ensures that you aren't needlessly dithering the 24-bit portions of the file that are still 24-bit. The way Audition works: if you apply an edit to even one sample of a 24-bit file, Audition will consider the entire file / workspace 32bfp, even though you've only edited a single sample. What this means is that if you dither to 24-bit, you're (needlessly) dithering 24-bit to...you guessed it...24-bit. Hence, you may want to consider disabling dither in this instance. That said, if you're doing heavy editing, it makes sense to tick the dither checkbox, since you don't want to truncate if you've applied any substantial / meaningful edits.
<2> After highlighting the portion of the waveform you're editing, dither it back to 24-bit with a plugin. Then save the file as 24-bit with the dither checkbox unticked. The edited portion of the file has already been dithered from 32bfp to 24-bit, so no need to dither it again. And the rest of the file remains 24-bit, since you didn't edit it, so no reason to dither it, either. This method will work for any number or size of sections you wish to edit, as long as you only edit each section a single time. If you perform more than one edit to any portion of the file, you don't want to use this method, as you'd be editing, then dithering, then editing again, and dithering again - no point in dithering multiple times.