Note that I also see this at the office between a laptop and a
-current uniprocessor (kernel vintage yesterday).

ex0: flags=8863<UP,BROADCAST,NOTRAILERS,RUNNING,SIMPLEX,MULTICAST> mtu 1500
        capabilities=7<IP4CSUM,TCP4CSUM,UDP4CSUM>
        enabled=7<IP4CSUM,TCP4CSUM,UDP4CSUM>
        address: ...
        media: Ethernet 100baseTX full-duplex
        status: active

Different 100baseTX switch (Cisco this time).

There's no hang, but the ACKs are so delayed (some seconds) that
throughput is just awful.

However, sending from the same laptop to a linux box (eek) or
an Irix box works as expected.   Worse:

: netbsd-office ; scp -v irix-box:bigfile.avi .
...
Sending file modes: C0755 315031552 spiritual kf.avi
bigfile.avi       0% |                             |   418 KB 15:43:21 ETA
[wait a couple minutes]
bigfile.avi       0% |                             |   978 KB 15:19:53 ETA

very very very slow.  this is new in the past days, zero topology change,
zero change to the Irix box (in months in both cases).

cf. 

laptoy% scp -v irix-box:bigfile.avi .
...
Sending file modes: C0755 315031552 spiritual kf.avi
bigfile.avi     100% |                             |   702 MB 07:35

was -current's TCP borked yesterday?

(I note some commits; were these regression-tested at all, without
profiling, etc.?)

	Sean.