Grid doesn't have to be fancy or technically overwhelming. I say start small and start local. Commercial applications can be relatively small scale, and mostly employ current hardware resources. Groups such as the Enterprise Grid Alliance are promoting the concept of starting local. (David also talked about EGA in this commentary.)
A few months back, I had the opportunity to speak with Donald Deutsch, president of EGA, and vice president of standards strategy and architecture for Oracle Corp. He put it this way: "Were dealing with the here and now of grid computing. That is, using proven and standard available components, such as networks, application servers, databases, server computers, and storage subsystems." Good candidates for grid computing include "heavy-hit, transaction-oriented, its-got-to-be-there-or-youre-out-of-business applications ERP, CRM, general ledger, business intelligence," he said.
I've also had the opportunity to speak with IT managers that have had great success with limited internal deployments that encompassed a single department within their organizations. A grid can be even be contained within a chassis of five server blades, dynamically distributing or allocating workloads to available processors as required. Or, it can be dispersed across a network of PCs within a department.
Grid may replace clustering as the way to better distribute workloads and achieve higher levels of availability. The challenge is to determine the tipping point at which clustering is no longer as viable as going to grid.