An OS can game that kind of statistic, though, by spending more memory on what Windows would call working set (the "hot" set of pages for a process which are preferentially kept in physical memory) at the expense of the file system cache.
Also, predictive prefetching may increase paging in, but in an async fashion, so the application is never actually blocked. This may cause spurious increases in the stats. You need to examine the amount of time threads are actually blocked by the paging behaviour.
Also, predictive prefetching may increase paging in, but in an async fashion, so the application is never actually blocked. This may cause spurious increases in the stats. You need to examine the amount of time threads are actually blocked by the paging behaviour.