It's perhaps even more trivial than that: no one can effectively estimate competence.

That's what the paper suggests, I should have been more clear in my statement. Neither incompetent, highly competent, nor people in between can effectively judge competence.

I think your summary doesn't include the most interesting bit imho: the sign of the error is different (incompetent overestimate their competence, the competent underestimate).

I think that bias makes sense: if you have 0% competence it's hard to underestimate, and if you're at 100% competence it's hard to overestimate. So if we assume that they're truncated normal curves and translate them into a simple average, that's what we'll get.