Your whole team could be having a new discussion within a thread and you'd never know unless you participated in the thread previously or somebody @'s you. Compare that to more IRC-like discussion where each message is posted to the channel. Sure, you may end up with contextual issues reminiscent of the IRC days, but at least everything is discoverable.
In my experience, the main problem here was that people regularly lost sight that a conversation even was in a thread. So they'd post their response unthreaded right in line. Now the people in the thread view can't see it, but if you were in the main channel it still looked just fine (after all they were one after the other). If it was a nascent threaded conversation (like 2 or so posts only), often the thread was just abandoned and everything was done top level. For bigger threads where you caught after the fact it was a threaded, people would either delete and repost or would just duplicate their post over into the thread again (which means it appears twice in the main channel). The worst offenders were people who refused to thread so even if it was humming along fine they'd just post their response inline always (and repeatedly).
Don't know if I have a particular point there other than both had tradeoffs. I do find that the Slack thread updates disappearing seems to be less frequent of an issue and the threading problems in the alternative were near constant. I think Slack's issue would be largely a nonissue if viewing a thread "subscribed" you to it so you could see updates without needing to post. Maybe reactions are a workaround too. It's fairly easy to unsubscribe if one's blowing you up. Subscribing to all threads regardless seems to reduce some of their usefulness and that's the current solution I believe.
That's kind of the point? You shouldn't be bothered unless you want to be. To me, a thread isn't necessarily supposed to be geared toward discoverability. You'll see the main topic hit the channel, and then you can choose to follow the thread if you like.