I bundle stuff up to make individual count idiosyncrasies easier to track and I leave them all as done_testing() without a count until I am done writing them and then I enter the counts. The counts RFC:SHOULD be there. There are cases where the tests will fail or skip and without the count, they might appear to pass. Simplistic example of my prefered approach–

Update: Added an idiom I use and like.

use strictures;
use Test::More;

subtest "This stuff with namespace and test count isolation" => sub {
    my @whatever = 1 .. 3;
    ok $_, "$_ is okay" for @whatever;
    return done_testing(scalar @whatever) if $ENV{FULL_MOON};
    ok "one more", "One more is okay";
    done_testing(1+@whatever);
};

subtest "That stuff" => sub {
    my %whatever = ( 1 => "One", 2 => "Two" );
    ok $_, "$whatever{$_} is okay" for keys %whatever;
    done_testing(scalar keys %whatever);
};

done_testing(2);
[download]