Basically I start with a color version, and process and sharpen normally. Then I duplicate the background layer, convert it to B&W, and do a bit of additional sharpening and curves to get the best possible B&W rendering of the image. Then I convert to LAB mode without flattening and hide the B&W layer. The last step is to apply the L channel of the B&W layer to the L channel of the color layer.
The tonality tweaks made during B&W conversion and processing get applied to the final color rendering, but without affecting hue or saturation. That's what gives the image the lazlo/dragon look.