Resolving a page fault requires loading the needed page into…

Resolving a page fault requires loading the needed page into memory from disk. This disk read operation dominates the time needed to resolve the page fault. The page read in from disk is placed in a frame; processes are given a set allocation of frames to use while they run. Why does resolving this page fault take twice as long if no ‘clean’ pages are found in the frames available to this process?