Bad benchmarks do not a good case make:

#! perl -slw
use strict;
use Benchmark qw[ cmpthese ];

our $fred = 'fred';
our $bill = undef;
cmpthese -1, {
   definedor => q[
       for( 1 .. 1e3 ) {
           $fred //= 'fred';
           $bill //= 'bill';;
       }
   ],
   unlessdefined => q[
       for( 1 .. 1e3 ) {
           $fred = 'fred' unless defined $fred;
           $bill = 'bill' unless defined $bill;
       }
   ],
   ternary => q[
       for( 1 .. 1e3 ) {
           $fred = defined $fred ? $fred : 'fred';
           $bill = defined $bill ? $bill : 'bill';
       }
   ],
};

__END__
C:\test>junk71
                Rate       ternary unlessdefined     definedor
ternary       3989/s            --          -45%          -53%
unlessdefined 7283/s           83%            --          -14%
definedor     8455/s          112%           16%            --
[download]

But as tux says, it's not about speed, it's about convenience, conciseness and readability.

Performance gain (when measured correctly), is simply a nice side effect.