I've been looking at the kernel source to check what this setting affects, and although I couldn't see the direct link between this value and the pages written, the comments and documentation says it does affect it (I'll just assume it is my limited comprehension of the code from seeing how it does), however this value also affects the number of pages that are read ahead of what is requested, which is really for performance on hard drives where seek time is an issue, so the fact reducing this value reduces the number of unnecessary extra reads from swap may improve performance (or it may be insignificant).

Another thing I noticed was that the kernel defaults this value to 3 (or 2 on systems with less than 16MB RAM) and there is a comment that setting it much higher than this causes other issues. While these other issues remain unspecified along with the exact value to trigger them, it may be that that values of 5 or 6 could trigger them.

In all, these changes have seemed to make my N900 more responsive and maybe after a couple of days use with these settings (except the tcp_timestamps since I don't think that is worth doing), I might tweak them one at a time to see which ones really have an effect, my first one will be to change the page_cluster to 3.