I think this is the problem. We don't seem to make a difference between
	recently-used pages for a file we are writing, and recently-used pages
	by text or data process. When writing a large file I guess the most recently
	used pages are the ones for the file's data, and the pagedaemon free pages of
	programs instead. Maybe the pagedaemon should first free pages of file
	data ?

That's also done by my patch.  (At least that's what I thought when I wrote
the patch.)  Pages are immediately cleaned and put onto the idle list.  Of
course this implementation could cause other performance problems, like 
multiple writes of the same page, and does chew up lot's of CPU time.

Eduardo