The January issue of Communications of the ACM contains two very interesting articles about parallel DBMS and MapReduce: MapReduce and Parallel DBMSs: Friends or Foes? is written by several database guys and describes how a typical MapReduce program can be executed on a parallel DBMS. It also claims that the DB is faster for problems typically tackled with MapReduce.

The second paper MapReduce: A Flexible Data Processing Tool
written by the inventors of MapReduce assesses the original version of the first paper and claims that MapReduce is usually much more efficient than demonstrated in the benchmarks of the first paper.

I personally found the second paper more convincing but I´m by no means an expert in this area.