Windows 2008 R2 / ESXi 5.5 - 10 GBit Network Performance

Question

Hallo,
maybe my problem is by design, but I hope anybody has a solution to improve the transmission speed between Windows 2008 R2 Guests.
In my Test-Environment I have 2 Windows 2008 R2 Guests on 2 ESXi Hosts with Intel 10 GBit Nics which are Cross-Connected.
Between these VMs the maximimum transmission rate with iperf is 2.1 Gbits/s (Debian: 9.28  Gbits/s).
Iperf windows Setup
Server
iperf -s -P 0 -i 1 -p 5001 -f g
Client
iperf -c <SERVER> -P 1 -i 1 -p 5001 -f g -t 10 –w 128000
While testing with NetIO the maximum transmission rate was 397 MByte/s (with 32k packets).
ESXi Configuration
Version: 5.5 Build 1623387
NIC: Intel 82599EB
driver: ixgbe
version: 3.7.13.7.14iov-NAPI
firmware-version: 0x8000014d
bus-info: 0000:81:00.1
PowerPolicy: BIOS OS Controlled / ESXi High Perfomance
ESXi TSO / LSO
~ # esxcli system settings advanced list -o /Net/UseHwTSO
   Path: /Net/UseHwTSO
   Type: integer
   Int Value: 1
~ # esxcli system settings advanced list -o /Net/Vmxnet3HwLRO
   Path: /Net/Vmxnet3HwLRO
   Type: integer
   Int Value: 1
Windows 2008 R2 Configuration
Windows 2008 R2 SP1 with following Patches:
2519736
2525390
2511305
2675785
2775511
(Patch selection from this KB: http://support.microsoft.com/kb/2643970)
4 vCPUs
8 GB RAM (Full reserved)
VMXNET3
vHW 10
PowerControl HighPerformance
netsh Configuration
netsh int tcp set global rss=enabled
netsh int tcp set global chimney=Enabled
netsh int tcp set global autotuninglevel=normal
netsh int tcp set global congestionprovider=ctcp
netsh int tcp set global ecncapability=Enabled
netsh int ip set global taskoffload=Enabled
netsh int tcp set global timestamps=Enabled
VMXNET3 Configuration
RSS enabled
RSS works in my Tests on 2 vCPUs
To verify my results I installed 2 Debian 7.6 Guests with open-VM-Tools and VMXNET3 vNIC
Test Results:
Iperf: 9.28 Gbits/s (with default windows Size)
NetIO: 430 MByte/s (with 32k Packets)
To verify my results I installed 2 Windows 2012 R2 (without Patches) VMXNET3 vNIC
Test Results:
Iperf:2.3 Gbits/s (with 128k windows Size)
NetIO: 420 MByte/s (with 16k Packets)

Answer 1

Well, if you could reach 8.9 Gbits/s with these settings then I guess it's alright and I should rather say iperf requires a lot more custom tuning of options to reach a similar level of efficiency under Windows compared to it's Linux counterpart, which barely requires any custom options for maximum throughput.
I have tried the psping utility and it achieved better results than untuned iperf tests, though results were mixed overall:
http://technet.microsoft.com/en-us/sysinternals/jj729731.aspx