The bug is that you sort the second type position numerically instead of lexicographically. However, i don't like the way you leave the elements in @data as array references when you're through sorting. You should have finished up the Schwartz Transform so that the original data was still there, just reorganized. If we make those corrections we come up with:

my @data =
        map { join '', @$_[0..3],'R',@$_[4..6],'.',$_->[7], "\n" }
        sort { $a->[7] cmp $b->[7] ||
               $b->[6] <=> $a->[6] ||
               $a->[0] <=> $b->[0] ||
               $a->[1] cmp $b->[1] ||
               $a->[2] <=> $b->[2] ||
               $a->[4] <=> $b->[4] || # update: changed (see 1)
               $a->[3] cmp $b->[3] }
        map { [ /^(\d+)(\D+)(\d+)(\D*)R(\d+)(B?)(\d*)\.(\w+)$/ ] }
        <DATA>;
[download]

update: Unfortunately, before i even got to post this, Ovid posted a better solution that makes sure all the values are defined before being sorted as well.

1: this was changed due to Ovid's note...