> I ran a simple loop doing basically nothing

It sounds like you were measuring “ALS.run vs. doing nothing” performance, which is a different story than ALS overhead. It’s better to measure the overhead within a web application, the closer to what you have in real-world, the better it is.