Disagree about scripting.

Compile times are fast if you are willing to tradeoff performance. But you can decide whether you whether or not you want a long compile fast program, a short compile slow program, or an interactive compiler.

The example I can give is StandardML. You have SML/NJ and MLTon as two possible compilation options among other. SML/NJ has an interactive mode. MLTon compiles for longer but results in faster programs. (And it's also Hindley-Milner)

Just going to the "it should be interpretable" route without there being any option for whole-program optimization at compile time results in unnecessary bloats in performance which by all means we can avoid now as development hardware has become faster.