The main inefficiency is that bit operations are not really bit operations; rather, the module uses ASCII strings of 0s and 1s, and only packs it into a real bit string after the encoded string has been constructed. ...

On the encoding side, it is only slightly harder to build the bit vector directly, using vec() instead of pack 'b'.

sub huff1 {
    my ( $string, $codes) = @_;

    my ( $ret, $len) = ( '', 0); # $len counts bits
    for ( split( //, $string), '_eos' ) {
        vec( $ret, $len ++, 1) = $_ for split //, $codes->{ $_};
    }
    return $ret;
}
[download]

The decoding side is harder.

Update: No, it isn't. Dehuff is as easily adapted to vec().

sub dehuff1 {
    my ( $string, $code) = @_;
    my %decode = reverse %$code;
    my ( $ret, $c, $where) = ( '', '', 0);
    while ( 1 ) {
        $c .= vec( $string, $where ++, 1);
        next unless exists $decode{ $c};
        last if $decode{ $c} eq '_eos';
        $ret .= $decode{ $c};
        $c = '';
    }
    return $ret;
}
[download]

Anno