These kinds of questions are always best answered using a benchmark ;)

#!/usr/bin/perl -w
use strict;
use Benchmark qw(cmpthese);

sub returnhashref {
 my %a=();
 for (1 .. 1000) {
   $a{$_}=$_;
 }
 return \%a;
}

cmpthese(-3,{
        'Ref' => sub { my $a=returnhashref(); return $$a{245}},
        'newhash' => sub { my $a=returnhashref(); my %a = %$a; return 
+$a{245}}});
[download]

Benchmark: running Ref, newhash, each for at least 3 CPU seconds...
       Ref:  3 wallclock secs ( 3.24 usr +  0.01 sys =  3.25 CPU) @ 38
+1.54/s (n=1240)
   newhash:  3 wallclock secs ( 3.21 usr +  0.00 sys =  3.21 CPU) @ 21
+1.84/s (n=680)
         Rate newhash     Ref
newhash 212/s      --    -44%
Ref     382/s     80%      --
[download]

So using the ref is about 50% faster in this case. Btw. the easiest way of dereferencing is using the shortcut $$a{somekey} instead of $a->{somekey}

T I M T O W T D I