We've thought about a tree-based model, but our current thinking is that a feature of that form would add a lot of UI complexity, for fairly limited benefit. One of the key things we'd lose with a tree-based model is the ability for someone catching up on a conversation to efficiently read it without a lot of jumping around between different views (or having to read a tree, rather than a linear history).

With Zulip's existing model, you can always fork off a side discussion by just starting a new topic, if appropriate adding links in one or both directions (and people pretty naturally do that when it makes sense to). Some Zulip users are fans of a convention to name a topic "original_topic.d" for a "digression" off a topic, but I think it's a much better approach to choose a topic that more directly suggests what the forked conversation is about.

(And I agree search is really important! If you're having lots of high-value conversations in a tool, it needs to be easy to find them later. Zulip's powerful full-text search has been praised as a lot better than Slack's since our very first prototype version of it. I think a big part of the reason is that the threads provide the context to make it easy to find the rest of a conversation. But we're always working on making it better).