Having not writ none, nor having ability to do same, pinch of salts are a given
but it read to me, as one of the serial-complainers of this issue over time, that
we seem to have either the wrong range of hammers to hit this screw, or else
don't yet understand the problem well enough to even agree fixes have merit.

I know you (who do write) hate it when we nag "it feels slow" 

But can you agree that the heuristic(s) around choosing what to do, when an
apparent resource is close to exhaustion under sysctl/vm tuning, don't quite
seem to fit the current situations?

I try diddling the min and max. Its a black art. Perry did some for me, he managed
to convince me to back off and let the machine do its own.

I got told to go into NEW_BUFQ. I'm not convinced I can tell a meaningful 
difference before or after.

nice doesn't have global coverage over this. nice <low> stuff can stop either
its peer <low>ness or even higher 'ness in this space.

once you've walked your filespace via stat (cvs?) it can be a one-way street
until something significant (like shedding a 60-100mb process or two) happens,
which reeks of TOPS-10 shunting paged programs out, and forcing loss of COW
space and the like. (MUD players used to force this by running every Galaxy
command in the book to wipe the 'non-guaranteed' DB held state out of memory
and force a reset of game state on the original tops-10 one off the DECUS tapes)

If you don't suffer this, if you think the sufferage is minor pain compared to
other things in NetBSD, I can't complain or realistically expect to change your
mind with such a piss-weak case, but I think all of us on the 'client' side
would probably back down a bit if we got the loose sense that you did also
see this.

So, can I ask the $64 question?

do core(ish) people who run their own dogfood not see this? if you do, but
just don't care *tm* I won't hate you, or have a tantrum.  But a lot of the
silence reads as 'nothing to see here, move along' when I honestly believe
its a bigger issue than that. 

no, its not going to stop me running NetBSD. 

-George