This page gives a nice overview of some common Map Reduce patterns in a broad range of applications from creating an inverted index, to implementing Page Rank and doing joins and other set operations
http://highlyscalable.wordpress.com/2012/02/01/mapreduce-patterns/
Basically wherever you see linear algebra (matrix/vector operations) you can apply Map Reduce. This means there are many applications in collaborative filtering (think Netflix recommendations), clustering, and other machine learning algorithms. Another very interesting project to look into is Mahout, which aims at implementing scalable machine learning algorithms, to a large extend on MapReduce through Hadoop.
http://mahout.apache.org/