Not related to performance, but rather at ease of use during programming. When I have a subroutine that does something to the values it is passed, I have it run different code paths depending on how the subroutine is called. For example:

sub foo {
    if (defined wantarray) { # we're in scalar or list context
        my @param = @_;      # make a copy of the parameters
        foo( @param );       # recursive call for ease of maintenance
        return @param;       # return the result 
    }

# we only get here if in void context
# do what you need to do on @_ directly, only thing to maintain

} #foo
[download]

This allows you to use the subroutine in two ways. One directly modifying the parameters passed:

foo( @data );
[download]

and one returning the changed parameters, keeping the old:

@newdata = foo( @olddata );
[download]

This gives me a lot of flexibility: on the one hand the direct changing which is CPU and memory efficient (no need to copy anything) and the copying way, which also comes in handy if you're passing constant (non left-value) parameters, like so:

@data = foo( qw(foo bar baz) );
[download]

If you're really concerned about performance, you could remove the recursive call, but that would leave you with two identical code paths to maintain, which is always a bad thing.

Liz