Thanks for your comment! And thank you for pointing out the raise_error switch - which I wasn't aware of (rtfm...)!

Running your test script gives me these errors:

test: http://cn.life.dada.net/people/
is decoded content utf8? 1

test: http://www.sina.com.cn/
decode failed: euc-cn "\x82" does not map to Unicode at /usr/lib/perl/
+5.8/Encode.pm line 166.
 at test_ku6.pl line 22

test: http://www.ku6.com/show/34D6sgY4X6w3YegR.html
decode failed: utf8 "\xE6" does not map to Unicode at /usr/lib/perl/5.
+8/Encode.pm line 166.
 at test_ku6.pl line 22

test: http://www.xinhua.cn/
decode failed: euc-cn "\xA9" does not map to Unicode at /usr/lib/perl/
+5.8/Encode.pm line 166.
 at test_ku6.pl line 22
[download]

which points me at the problem that decoded_content uses the wrong charset for decoding. (right?)

A search on the issue led me to this helpful thread http://www.issociate.de/board/post/400895/Fetching_the_charset_when_set_in_meta..html and the non-cpan module HTTP::Response::Charset (which I haven't tried so far).

So I am stuck here with three options:
1. use Encode::Guess
2. use HTML::Encoding
3. use HTTP::Response::Charset

A quick test seems to show that HTML::Encoding is the most reliable (and much less a hassle to install than Encode::Detect). Now I will try

  my $enco = encoding_from_http_message($resp);
  my $utf8 = decode($enco => $resp->content);
[download]

But how do I combine it elegantly with LWP? (I am not a great module guru..)
Is there a way to pull it into the LWP context and get, let's say, $response->as_utf8 (which is the result of the above)?