I came across this old article:
...and I'm having a hard time conceptualizing why the "noise currents" from their CPU Vss/Vcc pins would "rush all the way to the corner of the board" -- to the bulk of their bypass caps -- unless the bypassing of the CPU itself was pretty awful to begin with. (Which it sort of sounds like, given there "when we hooked up a few dedicated bypass caps between the [really noisy Vss/Vcc] pairs, the noise currents... shrank...)
I do agree that "1uF cap per digital IC" is an arbitrary standard -- often you can get away with fewer caps! Although for something like a CPU I'd be tempted to stick some bypassing on each side. "Not breaking planes" is claimed to be "the biggest enemy," but I think in general it's good advice. They conclude by saying they're gonig to break up their Vcc plane into islands separated with ferrite beads between it, which seems as though it might hurt as much as it could help -- I'd be tempted to leave both the Vss and Vcc planes intract, and use Vcc plane -> ferrite bead -> bypass caps -> routed trace to Vcc pins on CPU instead.
Opinions?
---Joel