[DennisP]

I should have specified "for sorting algorithms that depend on comparing elements to each other."

You're right that it's difficult to measure algorithm progress, but using problem sizes that you're interested in running in practice is probably a decent heuristic. "I wanted to run this problem but it would have taken 80 years. But then I caught up on recent published algorithms and look, I can do it in an hour on the same machine."

I don't think I claimed the progress was steady. Then again, Moore's Law might get more jumpy as silicon wafers reach their limits, and we transition to memristors, spin devices, or whatever else they come up with.