The problem with most "traditional" RDBMSs is that they were not designed with modern web server environments in mind. In order to scale to handle the frequent and speed sensitive requests needed to serve web pages, techniques such as sharding or clusting are used to spread data across multiple DB instances. When you do this, you lose much of the traditional RDBMS functionality such as being able to join or order. Without this, how much do we really need SQL anymore?

It also makes for a scaling nightmare for startups who's #1 scaling issue is typically DB related.

In this respect traditional SQL is starting to show its age. My prediction is it will begin to become phased out for most web application purposes in favor of more easily scalable cloud-based and services such as SimpleDB and BigTable.