Awesome.  Not only is screen broken in this regard, but so is Savannah (it truncated my earlier comment).

Here it is again, reproduced entirely in ASCII.

I just ran into this issue, so I tried out the unicode++ branch.

Sure enough, code points such as <U+26ED9> come out correctly, and they seem to work in copy+paste, scrollback, writebuf, etc.

However, there are problems with the language tag characters (U+E0000 through U+E007F, as they get mangled by screen.  For example, if I output the text,

  <U+E0001><U+E007A><U+E0068><U+9089><U+E007F>,

screen turns it into a bunch of garbage that mutates as you scroll or move the copy cursor around.  The results differ every time the text is output, too.  This text disrupts screen's notion of where lines start and end: after printing this text, the copy mode cursor does not appear on the line that is actually being copied.

The 'unicode++' branch in the official repository [1] should now support plane-1 unicode stuff. Using this branch, I can visit, for example, http://www.code2000.net/oneplane.htm, and the characters do not show up as '?' any more.

If you (or anyone) could try it out, that'd be great. If someone's testing, please also test copy/paste from the scrollback, writebuf, hardcopy etc. commands.

[1] http://git.savannah.gnu.org/gitweb/?p=screen.git;a=shortlog;h=refs/heads/unicode%2B%2B