These are the results from my tests using 5.6.1. length seems to work fine for me whether use utf8/use bytes was inforce, but someone mentioned that there was a known problem with use bytes in 5.6 earlier in the thread, so I tried unpack with 'C*' which is cited in the docs as explicitely bypassing the unicode stuff.

#! perl -sw
use strict;
use LWP::Simple;
my $content = get( 'http://www.columbia.edu/kermit/utf8.html' );

{
    use utf8;
    my $c_len = length $content;
    my @c_bytes = unpack 'C*', $content;
    my @c_chars = unpack 'U*', $content;
    print "Charwise - length:$c_len; 'C*':", scalar @c_bytes, "; 'U*':
+", scalar @c_chars, $/;
}
{
    use bytes;
    my $b_len = length $content;
    my @b_bytes = unpack 'C*', $content;
    my @b_chars = unpack 'U*', $content;
    print "Bytewise - length:$b_len; 'C*':", scalar @b_bytes, "; 'U*':
+", scalar @b_chars, $/;
}
{
    open JUNK, '>', 'junk' or die $!;
    binmode(JUNK);
    print JUNK $content;
    close JUNK;
    print 'Actual (from os): ', -s 'junk', $/;
}
__END__
C:\test>239788
Charwise - length:31946; 'C*':31946; 'U*':28621
Bytewise - length:31946; 'C*':31946; 'U*':28621
Actual (from os): 31946
[download]