MJR> The charset= parameter tells the browser how numeric entities
MJR> are to be interpreted.

JWC> Not at all.  Numeric character references are always to be interpreted
JWC> as Unicode.  Some browsers are buggy about this, it's true, but
JWC> all 5.x-7.x browsers get this right.

Okay, my mistake.

So to return to my example: you can have a Latin-2 document which
includes literal <ż> (z overdot) characters as bytes of value 191,
but within that document &#191; still refers to the inverted question mark.
And a quick test verifies that this is the case.  It seems very odd to
see characters which don't exist in Latin-2 popping up when the
document encoding is set to Latin-2 . . . mental adjustment to this very
handy fact will no doubt take a little bit. :)

Thanks for the correction.