I take a inside-out approach and use CGI's distributive magic for building the td and tr blocks.

#!/usr/bin/perl 

use warnings;
use strict;

use CGI;

use constant COLS       => 4;
use constant ROWS       => 5;
use constant TABLE_SIZE => COLS * ROWS;

my $cgi = new CGI;

# Loop through the pages, to pretend we got a page number from a reque
+st...
foreach my $page_num ( 1 .. 6 ) {

  # Create a dummy list of image names. We're adding an orphan to test
+ a short
  # table display.
    my @image_names = map {"image_$_.jpg"} 1 .. 101;

    # ----------------------------------------
    # this does the real work...
    # ----------------------------------------

    # Get a page's table's worth of image names and wrap them inside <
+img>
    # tags...
    my @imgs = map { $cgi->img($_) }
        splice( @image_names, ( $page_num - 1 ) * TABLE_SIZE, TABLE_SI
+ZE );

  # Make sure the table has enough elements. Add blank elements if not
+ to keep
  # the rows balanced...
    if ( scalar(@imgs) % COLS ) {
        push @imgs, ('&nbsp;') x ( COLS - ( scalar(@imgs) % COLS ) );
    }

    # Read "THE DISTRIBUTIVE PROPERTY OF HTML SHORTCUTS" in the CGI pe
+rldoc.
    # 
    # Process rows in column-sized chunks...
    my @rows = ();
    while (@imgs) {
        push @rows, join( '', $cgi->td( [ splice( @imgs, 0, COLS ) ] )
+ );
    }

    print $cgi->table( $cgi->Tr( \@rows ) ), "\n\n";

    # ----------------------------------------
    # end o' working part
    # ----------------------------------------
}
[download]