Out of curiosity, how much better, typically, is the 'optimal' blending rule than blending with some random or 'reasonable' weights? Never really thought to ask myself that question...