Though there are ways to do it in a single pass, I found it a bit easier to think about if one first splits the list out into two and then sorts them individually. Note I'm using Sort::Key::Natural to sort the mixed values in a hopefully intuitive way (Update: you may note I added the test case "12c" to demonstrate the difference to a simple sort).

use warnings;
use strict;
use Test::More;
use List::Util qw/uniq/;
use Sort::Key::Natural qw/natsort/;

my @in  = ( "2bc", "1", "12c", "12", "21", "1", "1a", "2", "2", "2",
    "2", "3", "35", "31", "2b", "4", "42", "5", "51", "2ac",
    "52", "6", "7" );

my @expect = ( "1", "2", "3", "4", "5", "6", "7", "12", "21", "31",
    "35","42", "51", "52", "1a", "2ac", "2b", "2bc", "12c" );

my (@nums, @mixed);
/^\d+$/ and push @nums, $_ or push @mixed, $_ for @in;
my @out = uniq(
        sort( { $a <=> $b } @nums),
        natsort(@mixed),
    );

is_deeply \@out, \@expect or diag explain \@out;

done_testing;
[download]

As opposed to your previous thread, these tags look much more like HTML/XML, so you really should use an appropriate parser!