Not sure how sophisticated their traffic monitoring has become but to avoid this it used to be as simple as having the dyno ping itself. Doesn't avoid the hard cap on hourly compute though.
Do you think people working on tutorials that have them deploy a chapter's exercise to heroku, to never look at it again, are setting up things to have the dyno ping itself?