Yes, I agree that keywords should be non-hygienic. The problem with using :key in lambda/kw is that it looks like a keyword itself. This only matters if the identifiers are going to look like (or be) keywords themselves.
>> The syntax may still change a bit; John advocates (keyword-lambda (a b
>> &key c d) ...) instead of (keyword-lambda (a b (c d)) ...).
>
> I see I have yet to write down why. Two reasons:
>
> 1) The use of &key (I assume non-hygienically) is much more transparent to
> the human reader. Parentheses could mean anything at all
>
> 2) It's much easier to search documentation for "&key" than to search for
> parenthesis-grouping rules.
These are good reasons.
While &key is used in Common Lisp lambda-list syntax, I would prefer
:key which is used by Gauche, Sagittarius and STklos. Guile and Racket
also use #:key which has the colon. So :key is closer to current Scheme
practice.
The twist is that some Schemes read :key as a keyword object whereas
others read it as a symbol. This shouldn't be a problem with a powerful
macro system (syntax-case, explicit renaming, define-macro) where we can
call procedures to convert between keywords and symbols at will.
&key does not have the above problem.
I also worked from the assumption that &key or :key is non-hygienic. In
the case of :key it probably has to be, since in some implementations it
will be a keyword object and those are currently non-hygienic everywhere.