customMediaType-2: RFC 3023 flawed?

RFC 3023 recommends supplying a charset header for
both application/xml and text/xml.  I think the TAG and
W3C should take a slightly different stand.  I'm fine
with text/xml, except for 

(a) I'd strengthen it to MUST, since the default (US-ASCII)
    is almost certainly always wrong, and
(b) I'd give serious consideration to deprecating text/xml
    and text/*+xml

As regards application/xml, application/*+xml, I think 3023
is probably wrong and we should take stand that the server
SHOULD NOT send a charset header because 

(a) there's no transcoding, so the in-band signaling 
    mechanisms of XML work just fine, and so
(b) the recipient will have a much higher chance in almost
    every case than the server of getting the encoding
    right.

 -Tim