Yeah, I'm saying that this is one of the issues that brought the popularity of nosql databases. Constraints are expensive, and you can gain performance (while accepting greater risk) if you remove them.
But it's actually a little more complicated than I make it sound above. By removing things like unique constraints and foreign keys from the database layer, you also end up with a design that can much more easily scale out instead of up. Distributed database. Lots and lots of nodes. etc.
NoSQL databases have their place.
Traditional SQL databases do too.
Databases. When you have 500 million users, you can't afford unique constraints and foreign keys in the db layer anymore.
Is he just saying that fully featured RMDBs don't have the performance to handle that scale? It seems like in some applications those checks will be necessary, and often the db will be the most efficient place for them. Not everything makes sense as nosql. Maybe I'm missing the point though.