I like the "pit of success" analogy. In design (whether it's API or language, or even physical mechanisms), doing the right thing should be the easiest way to go. C++ has a narrow ridge of success, with cliffs not far from either side.
A more orthogonal, more powerful and even more controllable language, to me, would be some variant of Scheme or Lisp that let you produce fully typed ASTs or even target code for the places where you need absolute control over performance, but high level enough to focus on the business domain problems most of the time. C++ demonstrably simply can't climb very far in the abstraction stack; people don't write web apps in C++. That's the metric that matters in the end.