You can't really design a fast language -- certain language decisions influence the potential of a program to run quickly, but you're talking about taking a plaintext program description and ending up with something that runs quickly -- there's a metric crapton of magic that goes on between one and the other.
Ultimately, its my belief that for a new language to be successful today, it has to be compelling in a way that's not already cornered by existing languages. Its not enough to say "My language will be as fast as C++, just as general-purpose, but with a smaller, more-regular, and more-expressive syntax." Even if you achieved all those goals in spades, you will have only an "interesting" language -- one that can be admired for its achievements but still for which almost no one will be willing to retrain their programmers for, or use in production. All existing languages have momentum no matter how bad, your new language has effectively zero momentum no matter how good. Its exceedingly difficult to overcome that disparity without offering something more than a subjectively-better new-old-thing.
Any new language today that cannot find its pitch, and a voice strong enough to shout above the cacophony of other boisterous languages is doomed.