I really encourage you to Benchmark instead of assuming. You will get surprises somewhat often. And assuming a simple arithmetic op is a performance killer seems silly. It took 10 million iterations to even be able to get this to benchmark without complaining it wasn't enough to be sure; and while a fifteenth of a millionth of a second is indeed three times faster than a fifth of a millionth of a second… does it actually matter on any level for the application in question?

moo@cow[483]~>perl pm-dice
 use strictures;
use Benchmark "cmpthese";

sub d10 {
    my $self = shift;
    return rand(10);
}

sub d10_2 { rand(10) }

sub d10_3 { 1 + int( rand(10) ) }

sub d10_4 { 1 + rand(10) }

cmpthese( 10_000_000, {
    orig => \&d10,
    orig_simple => \&d10_2,
    int_plus_one => \&d10_3,
    plus_one => \&d10_4,
});
__END__
                  Rate         orig int_plus_one     plus_one  orig_si
+mple
orig          4048583/s           --         -19%         -38%        
+ -74%
int_plus_one  5000000/s          24%           --         -24%        
+ -68%
plus_one      6578947/s          62%          32%           --        
+ -59%
orig_simple  15873016/s         292%         217%         141%        
+   --
[download]