I just noticed that the implementation of the Color we pass to Spritebatch.Draw is different from the way it is implemented in XNA 4.0 : it appears that XNA passes the color to the effect using a single 4 byte parameter set (ps input?), while MG sets the color to each and every vertex. In a common scenario where we draw most of our images with Color.White the overhead in memory can be very significant… I can only see a performance gain in MG approach when Color is constantly changed for every draw call as we actually avoid stopping the batcher if only the color has changed…
Just my 2 cents obviously.