ASP.NET and NLB?

We have an application that we license that uses ASP.NET and the Windows
Network Load Balancing Services on a Windows 2003 network. We have 4
front-end web servers running W2K3 and using a NLB common IP address within
our Extranet internal network. We also have a "state" server that is
running the .NET State Service where all 4 front-end servers point their
session to the state services -- using memory only; no SQL state.

I'm not exactly sure how the ASP.NET application works, but the software
allows us to configure a user-defined text for the header of the ASP.NET
pages for that particular server. The setting appears to be saved on each
server... because each servers' text can be different.

One of the things I have recently done to help isolate performance issues,
etc., was to change the header text to include the server name in front...
this way when I user accessing the web farm, the ASP.NET page will display
the server that the user is on.

From my understanding of how this works, I would assume that the NLB is
redirecting the user to the same server for each request due to the affinity
masks being set. Therefore, the user's session is initially created on the
server that he initially accessing and consequently, every page thereafter
is coming from the same server???

The issue that I'm having is this, our users are saying that the response
time is slow when they access a particular server (server2, in this case)
and is acceptable when accessing server1, 3, and 4. They're basically
saying that there is something wrong with server2. This is the only user
complaining.

1) Can someone please explain to me how ASP.NET sessions work with NLB?
Again, I assume the first server is the server that creates the session,
even though it is stored on the state server for common session between all
serves. However, because of the NLB and the affinity mask, the users are
still hitting the same box for each request.

2) Does anyone know of a way I can measure the performance? We capture all
IIS counters within our logs. I see there is a page response time field.
Can this be used to retrieve useful information? I would assume that the
response times would be subjective, considering some of our internal
customers have more bandwidth than others... so some pages would obviously
be severed to them more quickly than other pages.

Opinions?