It definitely does -not- get optimized.

perl -e 'my %bz = (x => 2, y=> 3, z=>4); sub baz { my $x = \%bz }; sub foo { my $x = { x => 1, y => 2, z=>3 } } sub bar { my $x = {x=>shift, y=>shift, z=>shift} } use Benchmark; timethese(-1,{foo=>\&foo,bar=>\&bar,baz=>\&baz});'
Benchmark: running bar, baz, foo for at least 1 CPU seconds...
       bar:  2 wallclock secs ( 1.03 usr +  0.00 sys =  1.03 CPU) @ 607349.51/s (n=625570)
       baz:  1 wallclock secs ( 1.07 usr + -0.01 sys =  1.06 CPU) @ 3462279.25/s (n=3670016)
       foo:  2 wallclock secs ( 1.06 usr +  0.00 sys =  1.06 CPU) @ 618264.15/s (n=655360)

Note the decimal point.