In fact, the line directive parser could be a procedure that takes the
directive as a free-form string, and outputs a Scheme object (foo ...)
that corresponds to typing #!(foo ...).

So it would be like a macro expander, pre-processor, or translator, to
translate foreign directives into Scheme directives.

#! Encoding: EUC-JP

would be fed to something like:

(lambda (line)
   (let ((words (split-at-whitespace line)))
     (if (and (= 2 (length words))
              (string-ci=? "encoding:" (car words)))
         `(encoding (string->symbol ,(cadr words)))
         #f)))

and the result would be as if the file had:

#!(encoding EUC-JP)