They are both 100% identical, internally. Any speed difference is the result of system noise and not because anything was measured. If you wished to measure the difference in compilation speed, use string-eval. That's the only time anything different will happen. I'm not going to run this code because any actual difference is going to be so infinitessimal that it will be undectable above system noise.
use Benchmark;
$bar = '';
timethese( 0, {
doubles => sub { eval "$bar" },
singles => sub { eval '$bar' } } );