I could fix the full duplex problem thanks to your ifconfig trick
   but now I Still have the kernel message:
   
   hme0: status=8000<DTIMEXP>
   
   dozen of times... when I use the interface heavily
   I also have 30% of interrupts from top when I use the network heavilty
   I do not know if this is normal
   
   CPU states:  2.0% user,  0.0% nice, 64.0% system, 34.0% interrupt,  0.0%
   idle

how heavily is heavily?  what config of ultra10?  older sparcs can not
drive the hme at full speed, so it would not surprise me if an ultra10
doing full speed would use a significant portion of the CPU.
   
   what could it be  to generate the kernel message???

these appear for many people.  there may be a latent bug in the driver
or perhaps there are hardware revisions that have bugs we don't work
around yet.  in any case, you can edit your kernel and remove this
statement if you like, it doesn't indicate a serious error and there
has been talk of removing it from the master sources anyway.
   



.mrg.