But a worker has more capacity than a donor does to learn whether small probability failure modes prevail in practice, and can switch to a different job if he or she finds that such a failure mode prevails.

This part seems exactly wrong. When GiveWell or Giving What We Can change their recommendations based on new data or arguments and explain their reasoning, the donations switch rapidly and en masse. EA donations have very little inertia.

Building an organization in a specific field, accumulating field-specific human capital (experience, CV, education), these involve putting years of effort into a particular project or vision. If you later find out that cancer biology was a bad move and you think that renewable energy is more important, your years doing a PhD in that area are now substantially wasted. Careers have very high inertia and investment in cause-specific capital, while earning power is flexible and donations can be highly responsive to new inputs.

I acknowledge that Jobs is a cherry picked example, but I think that the general principle still holds.

It is highly cherry-picked from two directions. Jobs gave up most of his Apple stock so that he captured a relatively small share of Apple's recent rise, and he is generally believed to have had more irreplaceable impact on his company than virtually all CEOs (although still Apple stock did not plummet with his death).