to see the difference, the test should actually use KCode in both cases:

require 'jcode'
$KCODE = 'u'
require 'net/http'
rss = Net::HTTP.get(URI.parse('http://rss.wikio.fr/a_la_une.rss')) #some RSS feed with french accentuated characters

then acctentuated characters are OK in MRI, not in JRuby, sorry for the first test that wasn't correct.

My script:

require 'net/http'
rss = Net::HTTP.get(URI.parse('http://rss.wikio.fr/a_la_une.rss')) 
puts rss

And the following output resulted (among many rss entries): polytechnique de l'université

And this is without setting $KCODE at all. Can you produce a script that will actually output the wrong content? Perhaps you used "p" instead of "puts" and our inspect impl isn't doing what it ought?

rss = Net::HTTP.get(URI.parse('http://rss.wikio.fr/a_la_une.rss')) 

here what I get in my jirb console is without the accents. That's different from the irb console. But that's not important at all. May be only a console issue.
Then you were right puts rss just behave correctly.

The bug I had actually occurs later with 'rexml/document' when processing the rss:

require 'rexml/document'
xml = REXML::Document.new rss
puts xml

Here I'm very sure MRI, will show the french accents in xml while JRuby don't have them. Instead, I get "?" characters for every accentuated character like:
"Coupe de l'America report?e" with jruby instead of "Coupe de l'America reportée" with MRI

If you prefer a much smipler test to try against MRI and JRuby is:

require 'rexml/document'; REXML::Document.new "<node>héllo accentuated world!</node>"

Any idea? I don't know how would could reame the bug then.

require 'rexml/document'; puts REXML::Document.new "<node>héllo accentuated world!</node>"

Since puts seems to have its importance with JRuby. So any idea?

We'll have a look at it...

puts "héllo accentuated world!"

Ruby prints out what it should, JRuby does not.

class Test {
public static void main(String[] args) {
System.out.println(args[0]);
}
}

And run from the command line:

~/NetBeansProjects/jruby $ java -cp . Test 'h\303\251llo'
h?llo

I'll look into the issues with rexml to see if we're closer on those, but the UTF8-from-terminal problem seems like a larger Java issue.

Tom: If you're going to do the encoding work here, wait for Raphael to confirm the fix before resolving this. Otherwise, open a new bug for the encoding stuff.

Still, the console bug failing to read accentuead characters you type (unless specifying -Dfile.encoding=ISO88591 may be) still sucks a bit and is awkward.
A work arround to that last bug is to escape manually accentuated chars but this is only a workarround.

Still, may be we should close that REXML bug and open a new one only for the console UTF8 bug. I don't have admin right to do that, so it's up to you. Thanks for the constant progress!