Oddly, this does not sound like homework (;^). Okay, since the source data is exactly like what you show (though I presume you want to handle lots of cases with different keys and values in the hash, but hewing to the same rigid syntax), this ought to work:

use strict;

$/ = undef;
$_ = <DATA>;

my ($prefix,$objdef,$suffix) = ( /(.*?\$obj = \{)(.*?)(\}.*)/s );
my @hashdef = ( $objdef =~ /\s*([A-Z_]+)\s*=>\s*(\'.*?\')/g );

print $prefix;
for ( my $i=0; $i < scalar @hashdef; $i+=2 ) {
    print "\n\"$hashdef[$i]\"=>$hashdef[$i+1]";
}
print "\n$suffix";

__DATA__

package Stats;
sub new {
  my $pkg = shift;
  my $obj = {
    TOTAL_REPLIES      => '6373',     
    TOTAL_TOPICS       => '574',      
    TOTAL_MEMBERS      => '211',     
    LAST_REG_MEMBER_ID => '85-1039338924',
    LAST_REG_MEMBER_N  => 'Unlimited_Destroyer', 
    M_ONLINE_COUNT     => '25', 
    M_ONLINE_DATE      => '1037453854', 
  };  bless $obj, $pkg;
  return $obj;
}
1;


;
}
1;
[download]

I didn't try counting the cpu load on this, but it should be negligeable.

Your description made it sound like the stuff outside the curlies for the "$obj" assignment should be left unaltered, so the first regex match/assignment splits the whole thing into the initial, middle and final pieces, and the second regex match/assignment locates all the strings to be maintained from within the middle piece (putting these in a plain array to preserve order, in case that matters). Then print the first piece, loop over the sub-parts of the middle piece to print without those annoying spaces, and finish by printing the third piece.

update:oops, forgot that you wanted double quotes around the hash keys in the print-out. Also, if you don't want to print the "$prefix" and "$suffix" parts, just drop those two print statements, and put the "\n" at the end of the print statement in the for loop.