For anyone who's been wondering, "Does QoS really matter and improve Zooming latency?" ranthalas kindly provides a concept clarification: "This is a bit of a common misconception that is actually correct in most circumstances… let's address the 'bandwidth is never fully utilized'… for example, you have a 1Gbps link between two switches. According to graphs, this link never uses more than 200Mbps. No issues. However, in latency-sensitive applications, what you're seeing as a 'not even close to full link' is misleading. Think of any link as either fully utilized or not utilized. When a packet comes into a switch, if there are no other packets on the wire, it gets put on the wire. If there is another packet being put on the wire, it gets queued and then put on the wire. It's an all-or-nothing situation.
What QoS does in the case of latency-sensitive applications is to say: 'If this type of packet comes in, it needs to be put on the wire ahead of any other packets that are waiting.' So while the difference is likely milliseconds, in voice and video that matters. In this case, we're not using QoS to shape or police traffic [but] simply to assign priorities and force other traffic to get preferential treatment.
So, yes, even if your link is not fully utilized, QoS does make a difference, especially in voice and video applications. Even more so in a shared collision domain medium such as wireless."