There is a growing sense that SQL vs. MapReduce or NoSQL is a question of fitness of the tool for the purpose at hand.
If your problem is best solved by commodity hardware working in parallel, then NoSQL solutions may be the path to take.
I have seen that expressed in a number of ways but not with a lot of detail on what factors drive the choice one way or the other.
With enough detail, that could make both a very good guide and topic map for those faced with this sort of issue.
First noticed on Alex Popescu’s myNoSQL blog.