This is a difficult distinction to make, so I'll just explain my process...

1.) beginning with a mostly-finished piece, colors sorted and blocked out.

2.) using the 'magic wand' or 'lasso' tools, select and de-select whatever areas/colors I would like dithering to be placed.

3.) using a soft brush at low opacity, or perhaps a gradient tool, I paint the soft transitions in.

4.) using 'SAVE FOR WEB' (in PS CS2), experiment with all of those sliders and settings, until I like the outcome. I usually aim for lower thresholds of color.

5.) sometimes, I like how a portion of the image looks, or color was effected, but I prefer other settings/effects on other portions of the image. So, I will 'cut out' the parts of the image I liked, and Frankenstein together an ideal arrangement from disparate settings. Sometimes, I may want the BG to be very truncated limited color, while the figures retain more shades. So, there's a lot of fiddling around.

in the end, the final result isn't any 1 setting. I may 'start over' at step 3 multiple times. I usually have a vision in my mind for how I'd like the final result to look, and am tinkering until I reach it.