I guess that nested quasiquote patterns are supposed to behave as the
quasiquote syntax of R7RS (as in which unquotes are inserted literally
and which unquotes escape the quasiquote) .

However, I don't find anything about it in the specification yet.

This is also relevant for the main thesis in SRFI 200:

Given the pattern (quasiquote (quasiquote (unquote x))), the SRFI 204
matcher will probably match the pattern (quasiquote (unquote x))
literally. The SRFI 200-modified "FHD" matcher will, however, match
against a more general two-element list whose first element is
quasiquote and whose second element will be bound to x.

I haven't checked the Racket matcher.

Marc