Network card: e1000g (Intel PRO/1000 MT) did not working properly

Question

Hi to everyone!
I'm having some strange trouble.
All computers attached to one DLink gigabit switch
Server #1:
nf4-based motherboard, Intel PRO/1000 MT net card, onboard net card (forcedeth driver used) Ubuntu Linux 2.6.25 kernel version.
All just working.
Server #2:
nf6-based motherboard, same Intel net card, SunOS 5.11 snv_86.
All just working.
Server #3:
Same as server's #1 nf4-based motherboard. Same onboard card. Same SunOS 5.11 snv_86 as server #2 have installed.
Its work. BUT..
Intel card is not working as expected.
It is seen by system. It is properly up and running. It is sending packent and receiving packets. Ierrs=Oerrs=Collis=Queue=0.
But thru e1000g0 interface no one computer can reach this one, and from server #3 no one computer is seen thru e1000g0. Exept...
Except two old Sunray1 devices (attached to the same switch for testing purposes)! But even these devices seen only by arp requests, and is not pingable.
It is some mistery, but if I turn on jubmo frames, these rays got a picture from SRSS, testing on this sevrer! Not for long, heh, they hung up fast, but this is proof a physical connection :)
So, I can not disable onboard network card on Server #3 - server becoming unreachable :(
I know, those network cards must work on Solaris - as they did it on server #2. I know, those network cards work with my nf4-based motherboards - as they did it on server #1.
But I can't uderstand, whait is going on with e1000g driver on server #3!
P.S. Kernel driver e1000g settings are the same. All difference from servers #2 and #3 is: #2 had Intel netcard installed then Solaris was setting up, and #3 had Intel netcard installed later. All netcards are working (I trying to swap them from one computer to another - did not change anything).

Answer 1

*) #3 thinks it is sending, but a NIC/driver error prevents it.NIC is fully health. It is tested on another PC. Driver error - or driver's bad implementation - is the one I'm thinking.
*) #3 has a bad transmit pair of wires (would affect all communication)This patchcord works w/o problems on other PCs and even on onboard NIC of server #3.
*) Network problemsHeh :)
*) Firewall or other device is dropping the ARP repliesIt is some very stealthy and very selective firewall, affecting only external NICs :)
BTW, then cord is attached to NIC, it sends "who has this IP" requests, asking to be shure there will not be an IP conflict. No one seen this requests.
*) #1 has a NIC/driver error preventing it from seeing the reply.Only for send from #3? It is like a firewall on #3 :) And, there is more PCs, not only #1-#3. No one can see #3's replays (or asks), not only #1..
I'm repeating: this PC works in network now on onboard NIC w/o problems. But onboard NIC is only 100 Mb/s, and we need 1 Gb/s.
I was thinking, Solaris took wrong MAC for NICs - but not, ifconfig shows the same MACs as marked on NICs.
Even more, yesterday i plumbed e1000g1 - second Intel NIC in server #3. And I was surprised - it worked perfectly!
But I make "touch /reconfigure; init 6" - and after reboot, no one of two Intel NICs working again :( They both have the same behavior - no one can see packets they sends.)
I even try to restrict connection on 100 Mb/s by driver's .conf modify - it works, link become yellow instead of green (take a note, NIC is able to "speak" properly with switch!), but all other stay the same.
Strange, really strange behavior. More strange is a fact this NICs work on another instance of same SunOS, but on different motherboard, while the same MB with same NIC is working w/o problems under Ubuntu. Looks like some bug in Solaris driver.