Visual Studio Team System have a option to create a Webtest based on your activities in browser. The problem is that this logs only POST and GET and no background activity generated by JScript.
For that problem HTTPFiddler can generate from one or more traced sessions a Webtest file. All you have to do is to mark the requests in fiddler and save it as web test
From this Webtest you can create a load test to verify that your application have the needed performance. At that point I run into the next problem
How to specify the typical usage of a AJAX application?
If you uses features lik autocompleextender (from ajax futures), how often is a callback to a web service made? Its lot easier to figure out a step by step procedure to finalize a order process in a online shop if you only have HTML documents delivered.
Also the easiest case, a updatepanel with a few controls inside, gives the user more options to navigate through the form. So I expect several more postbacks from clients as with a classic ASP.NET application. Means you need more Servers and NLB to host a AJAX enabled web application. On the other side I expect less traffic.
I tried to figure out how effective AJAX postbacks compared to classic ASP.NET postacks are. Excuse that, but my scenario uses a page with 2 buttons displaying the actual time. One is in updatepanel and one button is outside. The test clicks in one run on the inside and in the second run on the outside button. My expectation is that a AJAX user loads the page and clicks ten times the button ( what a nonsense, but I need that ).
The results shows no significant difference in various counters like requests/second. So my feeling (not shure at the moment) is that reendering of a AJAX page needs the same processor load as the same classic page.