Backing up your assumption and
kyle's comment, for the behavior you're describing (skipping pod coverage tests, etc.), the current recommended practice is to check an environment variable, specifically
$ENV{RELEASE_TESTING} and skip_all() based on whether it's true/false. I say 'current recommended practice' because there is no real standard, AFAIK.
I used to use
$ENV{AUTHOR_TESTING} in my distributions but
$ENV{RELEASE_TESTING} is the hot new thing decided upon during the
Oslo QA Hackathon.
Update: For quick .t examples, see the
Oslo QA Hackathon 2008/Achievements wiki page.