> 4) The de driver reports stuff like this before calming down:

> de0: abnormal interrupt: transmit underflow (raising TX threshold to 128|512)
> de0: abnormal interrupt: transmit underflow (raising TX threshold to 160|1024)
> de0: abnormal interrupt: transmit underflow (switching to store-and-forward mode)

I'm not really familiar with the fine print in that area, but I'll 
give a try...

The tranmsmit side of the Tulip has a FIFO.  After the FIFO has <N> 
bytes in it, the Tulip starts sending bits out the wire.  If N is 
less than the packet length and the DMA machine can keep up, everybody 
is happy and you have reduced the round trip latency. 

If the DMA machine can't keep up, the Tulip does something like fake  
a collision and try again.  And it gives an error to the driver so 
that N can be adjusted. 

There is a knob in the hardware that adjusts N.  Each time the driver 
notices an underrun error because the DMA side couldn't keep up, 
it kicks the knob up to the next higher setting.

So if your system is up for a long time, you should see messages 
similar to the ones above, but only a few of them.  After that, N 
should be big enough so that the hardware can keep up.  (One of the 
otpions is whole-packet.)