Let's take an instance called Instance A. Instance A is currently on the fediverse, which we'll say is pretty evenly distributed. No instance has a large enough portion of users whereby others would have problems with activity loss if they defederated, which is good. If any instance starts doing things that Instance A doesn't agree with, they can defederate, and less activity won't be much of a concern with defederating from that single instance.
But now, let's take Instance B. Instance B is planning to implement ActivityPub and join the fediverse, and when it does so, it will control 80% of the activity. In other words, it has as much activity as the rest of the fediverse combined.
However, Instance B isn't particularly trustworthy. They don't value the open web like the rest of the fediverse does, their moderation is extremely poor, and they haven't cared for general well being in the past if it meant raising profits.
Here, Instance A and instances like it have two options: defederate immediately, or wait and see.
If it defederates immediately, Instance A will see some users move to other parts of the fediverse because they're excited about the 5x increase in activity from Instance B. They probably won't go to Instance B now, but maybe Instance C or D. However, a lot of people will be fine. After all, activity is staying the way it is, and Instance B is untrustworthy anyway.
If it waits and sees, this allows people on Instance A to enjoy and get used to the 5x increase in activity. Not bad so far.
However, let's say Instance B starts having moderation issues (e.g., widespread hate speech and more-than-usual spam) as everyone reasonably predicted. Instance A now wants to defederate.
If it defederated before, no problem! Nothing needs to be done.
If it didn't and wants to start defederation now, good luck. Now, everyone on Instance A has gotten used to the 5x activity on Instance B, and you're going to have an extremely difficult time convincing them to cut the activity they see and the users they follow by 80%. Way more people will leave Instance A if it defederates now than if it had just defederated early on.
In other words, if people on Instance A come to rely on Instance B for the activity they're used to, way more people will join the camp of "I'm leaving if you defederate with Instance B" then if Instance A just defederated from the get-go.
Let's take another example. Instance B wants to try to grab a bunch of users, so after some time, they stop federating at all.
If Instance A defederated, the people there are fine. They never saw stuff from Instance B anyway.
If Instance A didn't defederate, then 80% of the content that people are used to will suddenly be gone. Most of the accounts they follow will be disconnected, and activity will fall a ton. These users on Instance A will have two options: stay, with a horrendous drop in activity and no posts from the accounts they're most interested in; or just go to Instance B.
In either case, Instance B will be fine. Most interaction was between Instance B users, so this won't be that much of a deal. But for users on other instances that are used to seeing stuff from B, it'd be catastrophic.
In short, defederating immediately has much smaller consequences than trying to defederate when whoever you want to defederate from controls most of the activity that your users see.