Some pages of memory in your working set may be rarely used. If so, then allocating physical pages of memory may be a waste. The kernel may be able to get better performance by using those pages for the disk cache - decreasing latency by hitting the disk less.