High availability would be the most common case when you need to look somewhere else. And it's a serious distributed systems problem that requires pretty big architectural and cultural changes, you can't just optimize for it later.
My comment doesn't necessarily contradict the article. I only mean to point out that a general purpose "non-optimized" thing still fit a whole lot of companies out there, depending on their scale.
EDIT: Although I concede that yeah some DB tech makes it easier to configure HA systems, if you can live with the downsides.