Measuring time of computation is counting the number of primitive operations. I'd say a language is a collection of capabilities that play nicely. Given one language, we can create a second that is also RE by taking each instruction in the first and adding a nop, or calculate ack(10,10) and throws it away, or whatever. Any algorithm that the second language computes with at least one step runs in fewer primitive steps in the first language by design.

Beyond a theoretical consideration, in practice there is much more than preference providing the difference between languages. If they were the same, you shouldn't have any trouble implementing combinators in brainfuck.