Performance is perfectly reasonable with JSON, but there are many more ways to make it slow if you're not careful. Postgres has to read the entire JSON content even if you only access a small part of it, that alone can kill performance if your JSON blobs are large. The lack of true statistics can also be an issue with JSON columns. They're still easily fast enough for many purposes, they just have a few more caveats than plain columns you should be aware of.
There's some other aspects that are not that intuitive, but also not a big deal once you know them. For example you can use a jsonb_path_ops GIN index to speed up queries for arbitrary keys and values inside your JSONB column, but you need to use the @> operator so that the index can be used. If you write queries like WHERE myjson->>'foo' = 'bar' it can't use the index.
But writing analytycal queries is definitely more complicated so the cost of not having a "normal" data model adds up over time. To allow business analysts to work with the data I endend up creating a a bunch of views om the data without JSONB fields as a quick fix. So though the approach worked, I am not sure I would do it again.