Before I first saw BeOS running on a colleague's machine back in the ~mid-late 90s (same guy who introduced me to Python) I used an SGI Onyx Reality Engine machine [1] (roughly $250K computer back in the day) for molecular mechanics simulations and BeOS ran circles around it on perceived responsiveness. I really wish we have OS's that prioritize input/output latency over all else.
[1] https://en.wikipedia.org/wiki/SGI_Onyx