bug? mtu 9000 problems with iscsi target

I have the following setup:
iscsi-target build r1245, one target with two partitions, mounted on the initiator, plain kernel 2.6.12

open-iscsi r363, kernel latest Fedora  2.6.12-1.1372_FC3

both machines have an sk98lin driver for the network interface, both have the mtu set to 9000 (driver version 6.23), working in GBit mode full duplex.

When I try to copy a 450 MB file from one partition to the other (both on the target as seen from the initiator) everything works fine, until approx 40%, then the target gets error messages in the /var/log/messages, but no kernel panic and seems to continue:

Jul 23 11:22:09 indigo kernel: istd1: page allocation failure. order:2, mode:0x20
Jul 23 11:22:09 indigo kernel:  [<c01585cb>] __alloc_pages+0x2db/0x3e0
Jul 23 11:22:09 indigo kernel:  [<c015b8ac>] kmem_getpages+0x2c/0x80
Jul 23 11:22:09 indigo kernel:  [<c015c983>] cache_grow+0x133/0x380
Jul 23 11:22:09 indigo kernel:  [<c015cec6>] cache_alloc_refill+0x2f6/0x320
Jul 23 11:22:09 indigo kernel:  [<c03147e8>] sock_common_recvmsg+0x48/0x70
Jul 23 11:22:09 indigo kernel:  [<c015d2d2>] __kmalloc+0x72/0x80
Jul 23 11:22:09 indigo kernel:  [<c0314e61>] alloc_skb+0x41/0xf0
Jul 23 11:22:09 indigo kernel:  [<d0ae9722>] FillRxDescriptor+0x22/0xc0 [sk98lin]
Jul 23 11:22:09 indigo kernel:  [<d0ae9621>] FillRxRing+0x81/0x160 [sk98lin]
Jul 23 11:22:09 indigo kernel:  [<d0ae7d55>] SkGeIsrOnePort+0x155/0x3e0 [sk98lin]
Jul 23 11:22:09 indigo kernel:  [<c0231f76>] acpi_ev_sci_xrupt_handler+0x12/0x19
Jul 23 11:22:09 indigo kernel:  [<c0150bf2>] handle_IRQ_event+0x32/0x70
Jul 23 11:22:09 indigo kernel:  [<c0150d2c>] __do_IRQ+0xfc/0x320
Jul 23 11:22:09 indigo kernel:  [<c0367e08>] inet_ioctl+0x68/0x100
Jul 23 11:22:09 indigo kernel:  [<d0a287e0>] is_data_available+0x30/0x50 [iscsi_trgt]
Jul 23 11:22:09 indigo kernel:  [<c0105906>] do_IRQ+0x36/0x70
Jul 23 11:22:09 indigo kernel:  [<c0157ffe>] buffered_rmqueue+0xde/0x300
Jul 23 11:22:09 indigo kernel:  [<c0103c26>] common_interrupt+0x1a/0x20
Jul 23 11:22:09 indigo kernel:  [<d0a2b78b>] iet_data_ready+0x1b/0x40 [iscsi_trgt]
Jul 23 11:22:10 indigo kernel:  [<c034e4bb>] tcp_data_queue+0x97b/0xae0
Jul 23 11:22:10 indigo kernel:  [<c034c92b>] tcp_ack+0x6b/0x610
Jul 23 11:22:10 indigo kernel:  [<c034f56f>] tcp_rcv_established+0x3cf/0x770
Jul 23 11:22:10 indigo kernel:  [<c0358fb0>] tcp_v4_do_rcv+0xe0/0x100
Jul 23 11:22:10 indigo kernel:  [<c03597ce>] tcp_v4_rcv+0x7fe/0x960
Jul 23 11:22:10 indigo kernel:  [<c015cec6>] cache_alloc_refill+0x2f6/0x320
Jul 23 11:22:10 indigo kernel:  [<c015fe5f>] shrink_slab+0x7f/0x160
Jul 23 11:22:10 indigo kernel:  [<c033c5b6>] ip_local_deliver+0x96/0x260
Jul 23 11:22:10 indigo kernel:  [<c033cc28>] ip_rcv+0x308/0x4d0
Jul 23 11:22:10 indigo kernel:  [<c031c695>] netif_receive_skb+0x235/0x290
Jul 23 11:22:10 indigo kernel:  [<c031c76e>] process_backlog+0x7e/0x100
Jul 23 11:22:10 indigo kernel:  [<c031c8d2>] net_rx_action+0xe2/0x340
Jul 23 11:22:10 indigo kernel:  [<c0150da7>] __do_IRQ+0x177/0x320
Jul 23 11:22:10 indigo kernel:  [<d0a265a9>] send_r2t+0x169/0x180 [iscsi_trgt]
Jul 23 11:22:10 indigo kernel:  [<c0129a92>] __do_softirq+0x42/0xa0
Jul 23 11:22:10 indigo kernel:  [<c0129b17>] do_softirq+0x27/0x30
Jul 23 11:22:10 indigo kernel:  [<c010590b>] do_IRQ+0x3b/0x70
Jul 23 11:22:10 indigo kernel:  [<c0103c26>] common_interrupt+0x1a/0x20
Jul 23 11:22:10 indigo kernel:  [<c03143c1>] lock_sock+0x1/0x170
Jul 23 11:22:10 indigo kernel:  [<c0344356>] tcp_ioctl+0x86/0x170
Jul 23 11:22:10 indigo kernel:  [<c03442d0>] tcp_ioctl+0x0/0x170
Jul 23 11:22:10 indigo kernel:  [<c0367e08>] inet_ioctl+0x68/0x100
Jul 23 11:22:10 indigo kernel:  [<d0a287e0>] is_data_available+0x30/0x50 [iscsi_trgt]
Jul 23 11:22:11 indigo kernel:  [<d0a2981c>] process_io+0x2c/0xa0 [iscsi_trgt]
Jul 23 11:22:11 indigo kernel:  [<d0a29c67>] istd+0x1d7/0x320 [iscsi_trgt]
Jul 23 11:22:11 indigo kernel:  [<d0a29a90>] istd+0x0/0x320 [iscsi_trgt]
Jul 23 11:22:11 indigo kernel:  [<c0141046>] kthread+0x96/0xe0
Jul 23 11:22:11 indigo kernel:  [<c0140fb0>] kthread+0x0/0xe0
Jul 23 11:22:11 indigo kernel:  [<c010131d>] kernel_thread_helper+0x5/0x18

When I do the same thing with the mtu set to 1500 no errors occur.

Albert