If you think of a "directory tree" as a real tree gaph, the minimum number of mkpath() calls you need is exactly the number of leaves in that tree. find() does a depth-first search, so there's a quite efficient way to determine leaves:

my $p = './a';
my @paths = ();

find( sub {
        return unless -d;

        print "$File::Find::name in $File::Find::dir\n";

        unless( $p eq $File::Find::dir ){
                print "CH\n";
                push @paths => $p;
        }

        $p = $File::Find::name;

} , $p );

push @paths , $p;

print "--> $_\n" for @paths;
# this has been tested
[download]

The amount of saved mkpath() calls depends on the directory tree, it's still O(n) where n is the number of directories but that's only reached if the tree is super-flat (depth=1).

But why use mkpath at all? It ends up in n mkdir() calls anyways, why not do that yourself in the find-sub and save the time mkpath() needs to recognize already created directories? mkpath() is not usefull if you do directory traversal anyways. For find() uses depth-first search, it's guaranteed that the parent directory always exists.

find( sub {
   next unless -d;

   mkdir File::Spec->catpath( ... )
} , ' ... '
);
[download]

--
http://fruiture.de