I used the "create spatially balanced points" in Geostatistical Analyst in ArcGIS 10 to generate a large set of points with equal inclusion probability but a "spatially balanced" dispersal. I then buffered these using different buffer settings, with the idea of explicitly testing the influence of sampling unit size on population estimates (in my case, tree cover) using a fixed area sampled. This latter detail is key - I sum of the area for the sum of the selection of buffered points to be the same between comparisons. A toy example: what's the difference sampling 10 polygons 1 ha each in area (10 ha total) versus 100 polygons, each measuring 0.1 ha (again, 10 ha total). The point layer created by the tool qualitatively has the desired dispersion, but when you start buffering points, some start to overlap, reducing the total area (plus violating any possibility of independence between samples).
What I'm looking to for is an efficient way to find features that DO NOT intersect. I saw a post asking a similar question for PostGIS, but was looking for a way to approach the question using ArcGIS.