Unless your problem is I/O bound (you can't get it off the disks fast enough, or network bound -- transforming data to a worker nodes takes too long) using Hadoop is the wrong choose. CPU bound problems are better solved with Grid solutions that do a better job of scaling up (with in a single node) and scale out to multiple machines. Taking a step back, you should always ask your self if this can be done on a single machine, taking advantage of Moore's Law.