> Then I think we should consider all of those changes together.

Ok, here is a complete patch.

>> Would it be safe to drop NODISP in the new function?
>> I see that most calls of 'message' with 'sit-for'
>> don't use the NODISP arg.
>
> I don't see a need to remove it, as one caller that uses it is enough
> to justify it, and the price is hardly significant.

The problem is that in most calls the argument NODISP is nil.