The first thing I'd say is please put your array loops inside your read loop. You gain a lot of efficiency that way. The solution I came up with was (and I'm probably abusing some variables):

#!/bin/perl
use warnings;
use strict;
package main;

my @averages = ();
my $count = 0;
while (my $data = <DATA>) {
    my @cols = split /\,/, $data;
    $count++;
    if ( $count == 1 ) {
        @averages = @cols;
    }
    else {
        for( my $i = 1; $i < scalar @averages; $i++ ) {
        $averages[$i] += $cols[$i];
        }
    }
}

for( my $i = 1; $i < scalar @averages; $i++ ) {
    print $averages[$i]/$count, "\n";
}

__DATA__
IMAGINGNY,1.45,0.42,1.54,1.49,1.47,1.36,1.81,0.47,1.8,0.55,0.38 
JBSQLTST,1.29,1.09,1.13,1.88,1.11,1.44,1.25,1.23,1.05,1.39,1.61 
SNYCSQL,4.58,4.24,3.87,3.9,4.13,2.04,3.34,7.6,3.58,1.26,7.45 
Snynetsrv,26,26.34,24.59,26.46,26.24,26.14,32.35,31.77,31.77,29.92,26.
+59 
W32SPLMCOR01,8.27,13.23,7.73,8.85,9.15,13.95,0,0,0,0,0 
W32SDAPSCT01,3.07,3.14,2.97,3.28,21.65,54.23,3.16,3.02,3.26,2.77,3.40 
W32SDASALM01,1.22,1.3,1.11,0.92,1.57,1.06,1.01,0.87,0.93,2.26,0.91 
W32SMSCSD02,15.38,12,22.32,23.3,19.74,46.42,2.06,1.7,2.17,2.85,1.74 
W32SPLMCOR02,13.24,13.23,7.73,8.85,9.15,13.95,0,0,0,0,0
[download]

And the output is:

C:\Code>perl columns.pl
8.27777777777778
8.33222222222222
8.11
8.77
10.4677777777778
17.8433333333333
4.99777777777778
5.18444444444444
4.95111111111111
4.55555555555556
4.67555555555556
[download]

If you want to deal with irregular length CSV, you'll need some logic to handle the different counts per column. Something like a push of a $value, $count will be necessary.