The following shows the memory being re-used.

sub func1 { my $x; print(\$x, "\n"); return \$x }
sub func2 { my $x; print(\$x, "\n"); return }

push @a, func1() for 1..5; print("\n");
         func1() for 1..5; print("\n");
         func2() for 1..5; print("\n");
[download]

SCALAR(0x226d1c)
SCALAR(0x225ffc)
SCALAR(0x226cb0)
SCALAR(0x1830a6c)
SCALAR(0x1830a84)

SCALAR(0x1830a9c)
SCALAR(0x1830ab4)
SCALAR(0x1830a9c)
SCALAR(0x1830ab4)
SCALAR(0x1830a9c)

SCALAR(0x1830910)
SCALAR(0x1830910)
SCALAR(0x1830910)
SCALAR(0x1830910)
SCALAR(0x1830910)
[download]

Note that the implementation of lexicals actually differs a lot from your description, but your post is an accurate description of how lexicals should be perceived to work.