With the help of Alan Fry I could manage to get a fast solution like this

sub gettitle {
    use Fcntl;
    my $file = shift;
    local *IN;
    sysopen( IN, $file, O_RDONLY, 0 )
      or die "while reading: '$file'\n";
    read IN, my ($str), -s $file;
    close IN;

    my ($info_block) = ( $str =~ /\/Info\s(\d+)\s0\sR/ )
      or die "cannot get /Info paragraph\n";
    my $searchpos = -1;
    my $info_start;
    while (1) {
        $info_start =
          index( $str, "$info_block 0 obj",
            $searchpos + 1 );
        die "cannot get position of '$info_block 0 obj'\n"
          if $info_start < $searchpos + 1;
        last
          if (
            substr( $str, $info_start - 1, 1 ) =~
            /\015|\012/ );
        $searchpos = $info_start;
    }
    my $info_obj = substr( $str, $info_start,
        index( $str, ">>", $info_start ) - $info_start +
          2 );
    my ($title) =
      ( $info_obj =~
          /\/Title\s*\(  ([^\015\012|\015|\012]*)  \)  /x )
      or return 'undefined';
    return $title;
}
[download]

I furthermore compared the performance of the above solution with Text::PDF and PDF-111 from CPAN. The test set consisted of 36 PDF files summing up to 3.8 MB.

runtime ratios of
index-solution-from-above : Text::PDF methods : PDF-111
were:
1 : 6 : 12

PDF-111 from CPAN has other flaws too. The author didn't respond to my questions. IMHO it should be dumped. It has a far to promiment place in the module hierarchy.