The same is true for the converse, as in both cases you get to be your own query planner and add client<->server latency on every join.

I wouldn't say it is as bad. Basically as you scale relational databases, you end up sharding and loose a lot of the relational features, and basically end up with NoSQL. In the majority of cases I see, people are not dealing with "big data" but think they are.

You can go really far with most relational database engines by adding hardware to a single instance instead of adding more instances.

> In the majority of cases I see, people are not dealing with "big data" but think they are.

Spot on. The amount of stuff you can do with a single $2,000 box these days is astounding.