Cohesive design vision, yes, but no apparent plan for expanding it until it's powerful enough to be useful, and it doesn't look interesting as a research language.
Any new language should have at least one use case where existing languages are inconvenient and the new language's design would make it a better fit.
As a language with a compiler, being faster than some of the slower interpreters out there is hardly a noteworthy goal. Sure it's faster than Ruby, but is it faster than Cython?
Quite possibly not. 'Everything is a (duck-typed) object' seems to call for dynamic dispatch even for basic operations on arrays and integers.
And 'everything in Snow can be changed', so congratulations, you just ruled out most of the caching that could mitigate this.
The idea of a boilerplate-free compiled language is a bit alluring, though realistically you either just hack something together in shell or Python, or it's big enough that you do want everything wrapped and scoped properly.