The reason why Slack ends up being prone to "noise" is because there's no sense of what each "sound" is about or its origin. Given the same number of messages, a Zulip-like model of putting messages in topics in channels makes it (hopefully) at least somewhat eaiser to differentiate between the sound for which you're listening and the other sounds about which you couldn't really care less.
If you have one of those voice-driven personal assistant gadgets (Echo, Google Home, etc.), try putting it between you and some other sound source (say, a TV) and give it a command. It'll probably hear you correctly, since it knows to listen in a particular direction and can differentiate between you and the TV. Now put yourself between the Echo and the TV and give the exact same command. The Echo probably won't even hear you, and if by some miracle it does, it'll get overwhelmed by all the noise coming at it. Same amount of data, but how that data is arranged matters significantly.
Same thing here. You're the Echo, the information you want is the user, and the information you don't want is the TV. Slack puts the user between the Echo and the TV, while Zulip claims to put the Echo between the user and the TV.