I suppose my rationale behind using join over string interpolation was that when working with directories, it is often advantageous to use an array instead of a string. This way you can resolve path components like '..' using pop instead of a regex. The other idea was that each piece supplied to the join would represent a single path component, which could be validated using a quick map for the extra paranoid. Of course, then $HOME would have to be split into something like @HOME.

Anyway, Benchmark says:

Benchmark: timing 1000000 iterations of interpolator, joiner...
interpolator:  5 wallclock secs ( 5.13 usr +  0.00 sys =  5.13 CPU) @ 
+194931.77/s (n=1000000)
    joiner:  4 wallclock secs ( 4.53 usr +  0.00 sys =  4.53 CPU) @ 22
+0750.55/s (n=1000000)
                 Rate interpolator       joiner
interpolator 194932/s           --         -12%
joiner       220751/s          13%           --
[download]

From:

#!/usr/bin/perl

my ($the,$fastest,$donut,$youve,$ever,$seen) = qw [ the fastest donut 
+you've ever seen ];

sub joiner
{
        my $x = join ('/', $the,$fastest,$donut,$youve,$ever,$seen);
}

sub interpolator
{
        my $x = "$the/$fastest/$donut/$youve/$ever/$seen";
}

use Benchmark qw [ cmpthese ];

cmpthese ( 1000000, {

        joiner       => \&joiner,
        interpolator => \&interpolator,
}
);
[download]