very cool on the fixes.  I wonder if the corruption one explains all
my client reintegration disasters.

I have noticed slow writes from my office machine to the server for a
long time, and never spent the time to figure it out.  It was the only
client I had a problem with - the others all have slower network
connections, and my nb tends to be in WD mode, even when local, so I
don't notice.  Things are much improved after upgrading rpc2 on the
server!

Time to write a megabyte went down drastically; now I get about 850
KB/s.  While not blazing (100 Mb/s Ethernet), this is with an old
server (P6-200 and over IPsec.  The server process took 3 seconds of
CPU time to process a 10 MB write; this is higher than I expected.

148 ~/%co/FOO > time dd if=/dev/zero bs=32k count=32 of=test1
32+0 records in
32+0 records out
1048576 bytes transferred in 0.011 secs (95325090 bytes/sec)

real    0m50.895s
user    0m0.000s
sys     0m0.019s

gdt 155 ~/%co/FOO > time dd if=/dev/zero bs=32k count=32 of=test2
32+0 records in
32+0 records out
1048576 bytes transferred in 0.010 secs (104857600 bytes/sec)

real    0m1.273s
user    0m0.000s
sys     0m0.012s

159 ~/%co/FOO > time dd if=/dev/zero bs=32k count=320 of=test5
320+0 records in
320+0 records out
10485760 bytes transferred in 0.250 secs (41943040 bytes/sec)

real    0m11.565s
user    0m0.001s
sys     0m0.099s