I think that the question is not complete at all. I mean, testing what with what purpose? When you want to benchmark the performance of a tool, you have to specify a concrete task to do. This output is usually compared with the performance for the same task but with another tool, i. e. ArcGIS Desktop vs QGIS or gvSIG.
A particular use case for testing could be running a terrain model for a particular region, reprojecting all points of a huge shapefile or doing some 3D rendering with flythroughs. Usually you count the time that it takes to do each task and then decide which tool is better.
So the thing is, choosing a task to do and see the performance, if not, you have to go to a general testing. Considering you do not want to do a general sw testing, what do you expect from ArcGIS for Desktop?