The main reason is what you say -- it harms readability. Much better to have functions behave like any other value, like any other variable -- and only be usable after they've been created.
I agree, although I also see some usefulness:
If your inner function names are sufficiently descriptive, you can write the high-level behaviour of the outer function at the top, first to be read, to get a quick overview of what it does. Then declare the inner functions below that (this also requires that the inner functions must not have surprising side-effects).
I've never seen or tried out code like that in practice, but it seems like it could work.
If I recall correctly the main reason why this feature is not supported by CoffeeScript is because of some awkward IE bug though.