Precisely my take too. Maybe we're even right?!

My sense is the crux is that you make explicit with CQRS is that you can offload reads to read replicas as long as you're OK with the read replicas being a bit more behind in the writes. Which gives you greater scalability like you say.