I would probably end up using RA3 if I were using redshift. The choice of single database with schemas vs multiple database comes down to your use case scenario. I personally don't see any need to separate databases (in MPP architecture) anymore especially with workload management. Previously, separating database and hardware was the only way to isolate compute, now you can manage it fine grained through WLM. Going from schema.object to database.schema.object doesn't really add any management efficiency. Neither does putting your raw data in a separate database allow you to take advantage of RA3 cold storage any better as things will likely refresh regularly.