Adding some overhead in the subroutine being called, the difference (if any) between the two approaches disappears altogether:

$ cat 645237.pl
use strict; use warnings;
use Benchmark qw/ cmpthese /;
use List::Util qw/ sum /;

sub foo {
  my ($volume, $message) = @_;
  my @foo = ($volume .. $message);
  @foo = map { $_ ** 2 } @foo;
  return sum(@foo);
}

sub bar {
  my $volume = shift;
  my $message = shift;
  my @foo = ($volume .. $message);
  @foo = map { $_ ** 2 } @foo;
  return sum(@foo);
}
cmpthese(100_000, {
        shift =>  sub { foo(1, 100) },
        inline =>  sub { bar(1, 100) },
        }
);

$ perl 645237.pl
         Rate inline  shift
inline 8396/s     --    -0%
shift  8403/s     0%     --
[download]

If your Benchmark shows your computer is able to perform an operation 3 million times pr. second, that's equivalent of a 1000 m2 highway sign saying: Don't optimize!

Update: Changed code to supply arguments to functions as pointed out by bruceb3.