it is not the most cost-effective way, but it's pretty good, to use a layer per supply. TI promoted this approach early in the fpPBGA game. If you use a low-impedance ground plane on top and bottom and one layer for each supply, the inductance is minimized. It's still not perfect, and, with the numerous layers you need just to route an escape from such a large BGA, the cost will give the boss heartburn, but all indications are that you'll get good results.

BTW, if you use automated assembly, 0201 cap's are pretty handy, though nearly impossible for hand assembly, as they fairly well fit the minimal spacing of the balls on fine-pitch plastic BGA's. That allows for placing bypass on the GND layers without interfering much with escape routing. Poured supply layers will give you the option of using fewer cap's, all of the same value, and small package cap's will allow you to place them anywhere you want without causing big problems.

RE