> I think losing first-class-ness is too much cost to pay.  Gauche heavily
> uses keyword arguments
> for implementation-specific extensions to ordinary procedures, and that's
> the merit of keyword
> arguments---you can add extra feature to existing procedures; if you need
> to define macros,
> you can define another procedure just as well.

Thanks for chiming in. This is a very useful perspective since you have
a lot of real-world experience working with them.

> Srfi-177 doesn't say
> keyword arguments
> should always be processed as if passed as rest arg, so an implementation is
> free to optimize
> when the caller provides keyword argument via keyword-call (as opposed to
> apply).

Indeed, for the implementations that have native keyword args, the macro
expansions are like this:

(lambda (:key a b c) ...)	; keyword-lambda
(foo :a 1 :b 2: c 3)		; keyword-call

So they are exactly equal in efficiency to implementation-native ones.
It's up to the implementation's static analyzer to find out whether it
can optimize the calls to use a fast-path that skips the generic keyword
parsing.