e.g. I can have "testobj" as an object, and also "testobj/child" as an object.
When downloaded to a regular filesystem, that breaks and you lose one object or the other object depending on which was downloaded first, because a regular filesystem can't do that.
Under the hood, the --dump function calls 'aws s3 sync s3://bucket ./bucket' so however aws-cli handles that.
Admittedly, the --dump function is the area that probably needs the most attention.
If you're just calling s3 sync, it will fail in this use case, because s3 sync does not account for this.
You'd have to get a listing first, parse it, and then figure out interactively with the user what to do with the conflicting object names.