From your input (not yet woven in ikegami's last post, working on it) I constructed this variation, which should replicate in a simplified form, what the routine in my more complex CGI::Application script will be like:

#!/usr/bin/perl -COE
use CGI;
use CGI::Carp qw(fatalsToBrowser);
use utf8;
use Encode;

my $cgi = new CGI;

my $formtext = decode("utf8", $cgi->param("formtext") ); # this may be
+ abc or contain chinese
my $filename = $cgi->param('file');

my $fh_in = $cgi->upload('file');
open(my $fh_out, '>', "/var/www/clipland/www.clipland.com/$filename") 
+or die $!;

binmode $fh_in;
binmode $fh_out;

local $/ = \4096;  # Don't wait to find "\n"
while (<$fh_in>) {
        print $fh_out $_;
}

$cgi->header(-charset => 'utf-8');
print $cgi->header();
print "$filename has been successfully uploaded... Now some umlauts: ä
+öü";
[download]

1. It only uses the -COE switch for Stdoutput and Stderr. (using -CS will give the error, despite ikegami's fixes) and as Juerd pointed out: it's safer.
2. I do not switch utf8 on for the Stdinput, thus I need to manually decode form data from text fields - right?

Is this OK? Anything I've overseen?

BTW: no kidding? I have to differentiate between text and binary data?? A great relief to hear that utf8 text files are also binary data... ;-)