Yes, there's a lot of fruit still to be plucked from running old algorithms faster. There's a night-and-day difference between an algorithm that runs just a bit slower than necessary and just a bit faster. A 7 day numerical weather forecast that takes 8 days to execute is not a forecast at all. A lot of important tasks include this crucial element of time. Correct algorithms that take too long are, for bounded-time tasks, not capable of correctly acting upon the world. Missing the deadline is the same as being incorrect.