If they all are compiled down to the same medium, why are some CLR languages generally faster (I know speed can be implementation dependent) than others?

One reason I can think of is the "compiled down" part....let's say we have a dynamically typed CLR language, the compiler of that language doesn't have the meta-data to make assumptions to optimize, etc., so the bytecode produced for same task as another language can be less efficient, even if they're of the same medium.