Cloud providers have globally facing services, so you hear of the problems. Dedicated server hosts have issues, but unless it directly impacts you, you never hear of it. Additionally, there's so much that PaaS providers do. On a dedicated host, if say, Passenger is misconfigured, that's your issue, not there's, so long as the server is still up and you can SSH in.

I think it's mostly because the dedicated servers at your host don't hide between the convenient API of a PaaS like Heroku. Much more things are happening when you do git push heroku than when just pushing on a remote of yours.

If you tried to write a few scripts to automatically deploy your code on N instances, manage http load balancing and all the nice but invisible features of Heroku, I'm sure you would be down more often than they are ;-)

We are at that weird point in the industry where computation gets commoditized and rented like it happened with electricity, but reliability is not improving.

It just seems like people are more willing to accept the higher amounts of downtime for platforms like Heroku and Amazon.

My dedicated server did go down today. I guess that's what I get for starting this comment thread.

Heroku's complexity is not just hardware, it is software: tons of software written in a variety of languages. It will have many more moving part that a typical dedicated hosting provider.