would an omniscient perfectly moral being scratch his/her/its butt?

(Side note: this conversation is taking a rather strange turn, but whatever.)

If its butt feels itchy, and it would prefer for its butt to not feel itchy, and the best way to make its butt not feel itchy is to scratch it, and there are no external moral consequences to its decision (like, say, someone threatening to kill 3^^^3 people iff it scratches its butt)... well, it's increasing its own utility by scratching its butt, isn't it? If it increases its own utility by doing so and doesn't decrease net utility elsewhere, then that's a net increase in utility. Scratch away, I say.

Second, you're assuming atomicity of actions and that's a bad assumption. In your world actions are very limited -- they can be done or not done, but they cannot be done partially, they cannot be slightly modified or just done in a few different ways.

Sure. I agree I did just handwave a lot of stuff with respect to what an "action" is... but would you agree that, conditional on having a good definition of "action", we can evaluate "actions" morally? (Moral by human standards, of course, not Pebblesorter standards.)

Third, you're assuming away the uncertainty of the future and that also is a bad assumption. Proper actions for an omniscient being can very well be different from proper actions for someone who has to face uncertainty with respect to consequences.

Agreed, but if you come up with a way to make good/moral decisions in the idealized situation of omniscience, you can generalize to uncertain situations simply by applying probability theory.

Fourth, for the great majority of dilemmas in life (e.g. "Should I take this job?", "Should I marry him/her?", "Should I buy a new phone?") the answer "what an omniscient moral being would choose" is perfectly useless.

Again, I agree... but then, knowledge of the Banach-Tarski paradox isn't of much use to most people.

The concept of CEV seems to me to be the direct equivalent of "God's will" -- handwaveable in any direction you wish while retaining enough vagueness to make specific discussions difficult or pretty much impossible. I think my biggest objection is to the "coherent" part while also having great doubts about the "extrapolated" part as well.

Fair enough. I don't have enough domain expertise to really analyze your position in depth, but at a glance, it seems reasonable.