I generally use Heroku for an alpha deployment, but then migrate to DigitalOcean for stable production deployment. I feel Heroku can get too expensive as the application scales up.
When you say "too expensive," I'm curious what you mean... Are you including time/labor in your cost-benefit analysis? For me, Heroku allows me to buy more engineering time (and less sys ops complexity) to devote to other problems. From this perspective, do you still consider DO cheaper? If so, at what ~number of dynos, does it become cheaper... (I realize this is an impossible number to pinpoint...)