>  > 
>  > I know Ragge recently mentioned nailing a vm related bug, and I think it
>  > was Chuck who noticed the large compile problems disappearing.  I'm not
>  > sure if this is related to the vm bug or not.  And then also, I'm not sure
>  > if the vm bug was fixed before the 1.4 release or if the fix is just in
>  > current.  Then thirdly, I've not yet compiled a 1.4 GENERIC+debug so I've
>  > not yet done any empirical data gathering.
>  > 
> 
> Did you find out about these problems? 
> 
The swap problem is analysed and the cause of the problem is well-known 
now. The result is that the pagedaemon deadlocks itself, and therefore
the system rather quickly runs out of pages. I have rewritten the page
table handling and I am currently debugging it (I found a bunch of other
problems while doing this).
Many thanks to Chuck Silvers that helped me to understand what was actually
wrong and pointed me in the correct direction for a good solution!

-- Ragge