As it turns out, having high-level abstractions means that in many ways it's easier to do automatic optimization. I look forward to automatic data-driven database tuning and query optimization.
If SQL optimization was as good as compilers I'd be all on this train, but I just don't think we're there yet.
So Malloy is akin to a language that compiles to C, which is then compiled to assembly. Does this seem like a worthwhile pursuit to you?
Personally, I don't think any SQL replacement will really take off unless someone can get the major RDMS makers to support it natively.
Just to be clear, I'm not saying it's not a worthwhile pursuit: SQL does have many problems and could do with a replacement that supports more modern approaches to problem solving. What I am saying is that unless someone comes up with something that can automatically do performance tuning for you, or that is supported natively by the RDMS, then SQL replacements will continue to fail to gain traction.