Indeed. There are a lot of applications which data is never traversed in a relational way. But there are other concerns to be raised on using a denormalized schema. Concurrency being the most important IMHO, database systems like PostgreSQL do a great deal of effort to behave nice and easy with a lot of concurrent access, but you can only benefit from it as long as your database is kindly normalized, otherwise there isn't much that can be done.
But I can't really say at which extent it matters, to which kind of application, at which volume of access. But at least as far as my experience goes, most of the time the bottlenecks are on the persistency, and there's a lot of science in making RDBMS scale.
Of course I'm not talking about the "solve the RDBMS scaling issue with memcached" paradigm.