Missing the point. (Score 3, Insightful)

The secret isn't getting a language/API to run faster than what's out there today. It's a matter of widespread adoption.

Just like BeOS was fast as hell, it didn't matter much because 5 people used it (sorry, Be people).

If you can create a language that will execute faster within the JVM, for example (hypothetical here), then you'd have something. Speed is relatively minor thing, and unless you code for compatibility, it won't matter how fast your stuff runs (or can be developed).