With Pipeline Pilot clustering, it seems that the improvements come from distributing jobs and creating sub-protocols that can be run parallel to each other, so that each job or sub-protocol runs start to finish on a single node, and processes that cannot be run in parallel, such as sorting, will not see as much benefit.

With Grid computing, it doesn't seem as clear. If my protocol needs to do sorting of a large data set, will I see performance improvements with a grid as opposed to a cluster (all else being equal)? Can a grid break up a job automatically, using nodes as if they were natural extensions of a single processor?