Well, is this a complaint about relational SQL databases? The only issues the author has that actually are tied to the SQL language are #2 (queries are largely unknown up front) and possibly #1 (small records with normalized relationships). Most of the other complaints have to do with ACID data consistency and the overhead it imposes in simple implementations (e.g. no row-level locking).
By leaving sql you're foregoing the ease of development aspects that it has. It's like leaving python to program in c. Memory is cheap these days, for anything but the largest websites it makes more sense to use a relational database with all data in memory (and this will be as performant as any other solution you create).
almost exactly the opposite. its much easier to just get all my data related to some 'entity' in some place then to have to gather it over a half a dozen columns with a weird syntax (I know SQL very well, I just hate it with a passion :)
