When this case came to me, the situation was that the Exchange Management tools (both the Shell and Console) were very slow to open.
Given that most of what we are doing when the tools first open is talking to Active Directory, I checked to see if other AD tools, like Active Directory Users & Computers, were also slow. In this case, we found that when any of the tools went against certain Domain Controllers, we did indeed see major slowdowns.
We gathered Network Monitor traces during the opening of the various tools, and the notable thing that we saw was that there were 5 second delays between many of our LDAP Requests and the corresponding responses from the Domain Controllers.
After much troubleshooting by our Directory Services team, including debug tracing of the AD processes, etc., that showed that AD performance was just fine, it was finally found that the problem was due to the EnableTCPA setting under HKLM\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters. The value was set to 1, enabling the feature, while the other Scalable Networking Pack (SNP) features (EnableTCPChimney, EnableRSS) were disabled.
According to the Windows Networking team this combination can cause the TCP driver on that machine to think that the sender has reduced its sending capacity. The TCP driver then begins to perform regular jobs in response to the low sending capacity, rather than just immediately responding to the requests. This behavior causes the slow response/pauses that we could see in the network traces. However, the fact that the TCP driver is waiting to send the outgoing packets is something that cannot be seen.
The solution: disable the feature by setting EnableTCPA to 0.