This morning I extended the test Bill linked to above (indirectly, same dog but different lighting/scene) with a shot at the tail of another roll.
That shot proved for me, for Delta 400 in DD-X at least, the point you make in your last sentence. I got nice detail where I wanted it from a frame incident metered and shot at 1600 but developed for EI 500. I'm going to try a frame or two shot at 3200 and developed at 500 on the next roll.
The camera meter though, center weighted and matrix, in my F5 was being seriously fooled by the white background telling me I'd be 2-stops over exposed, had I followed the camera's lead the shot would have lost the detail I wanted. (Good spot metering and zoning techniques would have worked fine too, they would simply have taken a bit more thought.)