Their measurements include invocation of the interpreter. And parsing TS involves bigger overhead than parsing JS.

But TS is compiled to JS so it's the same interpreter in both cases. If they're including the time for tsc in their benchmark then that's an even bigger WTF.