3d is always worse in terms of processing power. Complicated algorithums to decide what you can see round what, shadows of things, interactions between things, etc is complex, 2d you have flat things that are either in front of behind, and in terms of xna can be achieved by using levels for the drawing call so you draw in order from back to front the spritebatch does its "thing" and basically the framework ends up making 1 image which is sent to draw rather than tons of different bits.