In theoretical computer science, you're somewhat right. Constant reductions aren't as interesting as complexity reductions but still worth publishing papers on

In theoretical computer science, you're somewhat right. Constant reductions aren't as interesting as complexity reductions but still worth publishing papers on of course. In mathematics, it's common to get 10^9 constant speedups just be applying a clever method. In practical computer science/engineering like HPC, people are typically excited to get 10-20% speedups, because in a project with a million dollar compute budget, that'll pay their salary that year. A 3-4 speedup is almost unheard of for already-optimized codes like LAMMPS (although these numbers are a guess at the practical maximum speed), so it's probably newsworthy.