Using Perl 5.8.0 on a SuSE Linux system, I get "2" printed by both versions of the test (hard-coded wide character, and piping a 2-byte utf8 code into stdin).

In what sense have you made sure that stdin is actually receiving two bytes of character data (not counting the line termination)? I did it as follows:

perl -e 'binmode( STDOUT, ":utf8" ); print "\x{00A1}\n";' |
  perl -e 'use utf8;
    binmode(STDIN,":utf8");
    $x=<>;
    $x=~s/[\r\n+]//;
    use bytes;
    print length($x),$/;'
[download]

and this gave me the correct answer (2).