Swap lossage is an ugly feature of the netbsd vm system.  This has been
discussed several (many) times.  Apparently this was fixed in FreeBSD,
but in a 'bad' way.

A work around is to kill and restart the few processes which are being
piggy on your machine.  Here are my results on one of my machines.

--

atrium 4% uptime
10:01PM  up 11 days, 20:07, 8 users, load averages: 0.65, 0.46, 0.35
atrium 5% pstat -sk
Device      1K-blocks     Used    Avail Capacity  Type
/dev/wd0b       32760    17684    15076    54%    Interleaved
/dev/wd1b       32528    17972    14556    55%    Interleaved
Total           65288    35656    29632    55%
atrium 6% ps augx | grep -v USER | awk '{ vsz += $5;  rss += $6} END {print vsz " " rss}'
17172 4664
atrium 7% vmstat
 procs   memory     page                    disks         faults      cpu
 r b w   avm   fre  flt  re  pi  po  fr  sr ?0 ?1 ?2 ?3   in   sy  cs us sy id
 0 0 0 73992   412    8   2   7   0   0   2  0  0  0  0    0   19   7  1  3 96

This is on an i386 with 8 meg of ram.  Killing and restarting cron released
a meg of swap.  

root       110  0.0  4.1  4776  328 ??  Is   31Dec69   37:23.40 named 
Killing named released 7 meg.

root       119  0.0  0.0   312   44 ??  IWs  31Dec69    2:12.39 sendmail
Killing sendmail released 10 meg.

This brought me to within 3 meg of the statistics reported by ps.

pstat -sk
Device      1K-blocks     Used    Avail Capacity  Type
/dev/wd0b       32760     7536    25224    23%    Interleaved
/dev/wd1b       32528     8776    23752    27%    Interleaved
Total           65288    16312    48976    25%