Queue can also allow batching. It might be that you need to consume or pack or ship N items at a time, otherwise you're losing money. Trying to aim for minimal latency would literally hurt your business.
This article obviously has a very specific kind of workload/service/target in mind that they fail to define and they overgeneralize. This is bad advice in many situations that they failed to imagine.
This can be helpful for avoiding resource contention, or hitting an API limit.