The problem is that it isn't much better (if at all) in its current implementation. The current implementation allows tracking and identifying individual users (this may be or may not be fixed by Google, it has already been reported to their Github issue tracker). Since it is entirely based on your browsing patterns you can also be tracked cross device (people's browsing patterns rarely change).

For the moment 3p cookies are better, they can easily be erased, blocked or even isolated, and are restricted to a single browser.

The privacy of current tracking methods (3rd party cookies) is bad, which is why a lot of browser vendors are starting to block 3rd party cookies by default. FLoC may be marginally better than 3rd party cookies, but still browser vendors are mostly choosing to block it by default. There isn't a good reason to let either succeed.

Even if we assume that FLoC is entirely good, it's a false choice - why do we need _any_ tracking at all?

Yes, but it is unlikely to succeed.

The problem is that it is an unstable strategy. FLoC is strictly worse for privacy-conscious users if trackers don't change their strategy, and it is also strictly worse for trackers to stop using their current tracking techniques.

It is a bit like the prisoner dilemma.