[you] can most optimistically assume normal distribution of these traits in people in power
This is not maximally optimistic! We can hope we could come up with a system that (a) empowers unselfish people over selfish people and (b) protects the system itself against interference from the powerful. This is a difficult thing to achieve, and many have arguably failed, but that doesn't mean it isn't possible to do.
Centralized systems inherently offer more affordances of seizing power to selfish ends.
I think this is kind of unclear. If you do not deliberately engineer a government to manage the distribution of power, instead you will get an unmanaged distribution of power, which in particular will not obviously be well-placed to prevent an individual accumulating and then seizing power for themselves.
But even if true, I think I would still be in favour of central government because centralized systems inherently offer so many other things, which together are IMO worth it.
I don't think it does assume perfect follow-up, it just assumes roughly the same follow-up from them as you. I hear you that maybe people you tip into taking the pledge are systematically different in a way that makes you doubt that as well, but I'm not actually convinced this difference is that substantial.
Similarly, I don't think different amounts of income feels like a big problem with this sentiment to me, as long as their income isn't systematically less (or more!) than yours. It feels like an imprecision, but if it's true on average it's not one I particularly resent.
(I think the rest of your points seem fine so overall I still agree with your bottom line.)
To respond to a relatively small part of the post:
Individuals and institutions can be motivated to change their behaviour for the better on the basis of concern for others. (Otherwise, how could effective altruism be possible?)
Effective altruism works if only a few people can be unselfishly motivated in this way, whereas requiring unselfish motivation as a key part of the ordinary economic system that nearly everyone participates in is a much bigger ask, so I don't think it would be very surprising to believe that one worked and the other didn't.
Alternative career paths could include things like government policy roles that have say a tenth of the impact of a global hub EA policy role but still are the highest impact opportunities available
I'm not sure policy effectiveness works like this. Perhaps AI policy is particularly concentrated where AI innovation is happening, but I would have thought e.g. animal welfare policy is pretty global, and while bigger markets have more impact, they correspondingly have less tractability / more competition, so the smaller market may be easier to find success in. It's even more complicated than that when you realise that policies can be modelled on success in other countries, so positive change in Australia could be a useful resource for the global effort.
do you know if Marisa exclusively used they/them pronouns, or she/they, or what? I remember hearing something about this but I'm not certain and can't find any online profiles anymore :(