I dunno what you've expected but it's not a big surprise that two HW decoders have similar CPU consumption.
"default" settings are the best BUT:
1. I personally think 21th "SVP shader" is better than 13th (though this didn't affect performance in GPU mode)
2. Some believe "Decrease grid step" helps and in that case "by two with global refinement" is almost an ultimate value
So it's more like "highest" test in SVPmark