I used a different approach than most other replyers in the thread. Instead of so much matrix stuff I've abstracted away the ugly parts. Also, there's a clear separation between data structure creation and output formatting and output. This is less efficient, but it reads well IMHO and this is the style in which I prefer to code. It's easier for me to have refactored code like this. I, again, used my utility module ihb::List::Util and imported &group and &transpose, and produced the following code. (&transpose should be fairly obvious. &group is explained here).

use ihb::List::Util qw/ group transpose /;
use List::Util qw/ max sum /;

my @files = sort qw/
    B           Digest      Filter     MIME       SDBM_File  Time     
+util
    ByteLoader  DynaLoader  GDBM_File  NDBM_File  Safe       Unicode
    Cwd         Encode      I18N       ODBM_File  Socket     XS
    DB_File     Errno       IO         Opcode     Storable   attrs
    Data        Fcntl       IPC        POSIX      Sys        re
    Devel       File        List       PerlIO     Thread     threads
/;

my $width = 55;
my $padding = 2;

# Columnize
my (@cols, @widths);
for (my $rows = 1; 1; $rows++) {
    @cols = group($rows, @files);
    @{$cols[-1]} = grep defined, @{$cols[-1]};

    @widths = map max(map length, @$_), @cols;
    
    my $tw = sum(@widths) + $padding * (@widths - 1);
    last if $tw < $width or $rows == @files;
}

# Format columns
for (0 .. $#cols) {
    my $format = "%-$widths[$_]s";
    for (@{$cols[$_]}) {
        $_ = sprintf $format, $_;
    }
}

# Output rows
{
    local $" = ' ' x $padding;
    print "@$_\n" for transpose(@cols);
}
[download]

In the OP it's not clear how newlines should be handled if the outputted string is equal to the length, also, it's not clear how filenames longer than the width should be handled, so a newline is always appended.

ihb