The author's testing methodology appears unable to tell leakage from caching. If your machine isn't swapping, then any unused memory is wasted. Better to keep

The author's testing methodology appears unable to tell leakage from caching. If your machine isn't swapping, then any unused memory is wasted. Better to keep all sorts of speculatively cached data. For a web browser that includes downloaded resources, but it could also include the rendered bitmaps of page fragments, preparsed documents, or any manner of partial computation that may be reused later. Once there is memory pressure the code can start dropping these things.