Well yes, but for some organisations -- namely those without inhouse ops expertise -- it makes sense. I've done some work for a non-profit that pays over $3K per month to Heroku and it's probably worth it given the cost and risk to migrate off.

One also pays quite a high tax to use, say, Ruby vs, well, just about any other language implementation except for Python and non-HHVM PHP (see for example http://www.techempower.com/benchmarks/). Many would defend the use of Ruby as entirely justified on the grounds of productivity.

The point is, there are inefficiencies all over the place and most consider them not worth optimising out.

The blog post says that PX dynos are currently c1.xlarge (see https://blog.heroku.com/archives/2014/2/3/heroku-xl#performa..., paragraph 4).