Here's another way. It may be a better way. I've used Tie::File so that, even with very large input files, you won't run into any memory issues.

use strict;
use warnings;

use Tie::File;

my $infile = q{noa.txt};

my $wanted_re = qr{
    \A
    (?>
        (
            Pass \s \# |
            Elapsed \s Time \s+ : \s+ |
            CPU \s Time \s+ : \s+ |
            Virtual \s memory \s size \s+ : \s+ |
            Resident \s set \s size \s+ : \s+ |
            Major \s page \s faults \s+ : \s+
        )
        (
            [\d.]+
        )
    )
}msx;

my $last_stat_re = qr{ \A Major \s page \s faults \s+ : \s+ \z }msx;

tie my @indata, q{Tie::File}, $infile or die $!;

for my $line (@indata) {
    next if $line !~ $wanted_re;
    print $2;
    print $1 =~ $last_stat_re ? qq{\n} : q{, };
}

untie @indata;
[download]

I dummied up some additional input to what you provided:

$ cat noa.txt
...
Pass #123 ...
...
Elapsed Time : 1753.2 sec
CPU Time   :  753.2 sec
...
Virtual memory size  :  4472.6 MB
Resident set size    :  4362 MB
...
Major page faults  :  7153
...
Pass #the salt
...
Pass #124 ...
...
Elapsed Time : 9753.2 sec
CPU Time   :  953.2 sec
...
Virtual memory size  :  9472.6 MB
Resident Evil
Resident set size    :  9362 MB
...
Major page faults  :  9153
...
[download]

Here's the output:

$ noa.pl
123, 1753.2, 753.2, 4472.6, 4362, 7153
124, 9753.2, 953.2, 9472.6, 9362, 9153
[download]