You could see the 'Additional delay' under Client processing delay section for each request, under Advanced tab.
What happens if you make this delay to zero??
Does the play back show much difference in overall response time?
Generally, the time taken to load the entire page outside IBM Rational Performance Tester (RPT) cannot be mapped to the time taken to load the same page when invoked via RPT recorder. There are lot of parameters that RPT, as a load performance testing tool, accounts for (such as the traffic and connection information).
For Example: A page consists of several connection streams executing page elements in parallel where a page element is one part of the page such as an image or client-side script. Page element response time is the time from the first byte sent to last byte received. The response time for a page connection stream is the total of all of its page element response times plus any additional time establishing connections, if any. Page response time is the maximum response time of all page connection streams. Processing overhead time for data correlation, action scheduling delays under stress, Custom Code processing or HTTP processing are excluded from page response time. Each page element also has a delay associated with it. This delay is time that was observed while recording the page. If one page element has a direct dependence on the response from another page element, RPT honors this dependence. The additional delay may or may not be significant but RPT includes it by default in an attempt to behave as much like the browser as possible. It may very well not be significant and you can therefore reduce it or eliminate it.
Also, RPT processes these Client side delays in parallel or sequentially depending on how the application server returns them. Sometimes if you try modifying/disabling the client processing delay value, it may also disable the immediate transaction under Rational Performance Tester. Because, practically speaking, when you disable a request, you potentially invalidate some delays because in theory the disabled request could have been the basis for a delay. Therefore, RPT automatically recalculates the delays on the page when you disable a request. That is, if a later request used the disabled one as a base, the later delay request should be adjusted.
Now, let's come to your question about why such discrepancy between the response in the browser and RPT...
Often times when you see response times that are inaccurate it is because of recording your actions too fast and so you end up with more than one page being combined together. When recording, you need to be mindful and make sure to pause between mouse clicks. Do an action and wait 5 seconds before continuing on to the next action. You also need to pay attention to where your mouse is and make sure you have no "hover gifs" that you accidentally cause to be sent to the server while recording.
To see if you have these problems you have two things you can look at.
1. Open the test and click on the name of the test under test Contents in the tree. Click Select->Request. On the right hand side you will see a table of all requests in your test. Look at the delay column and look for really large delay values. If you have these, they are usually caused by recording too quickly, and act as an "embedded" think on the page. The Client Delay is supposed to simulate how long your client took to process the data from the server before sending the next piece of data. When you record too quickly, these values can be skewed. You can also go to a specific page that is taking too long and look at those specific requests. You can use the same Select button, or you can go to the Advanced tab of each request and look at its delay.
2. If you find the delays are your problem, then you can split the page where the delays are long since this was probably supposed to be two pages. This will move that delay to be the think time of the page which would be more accurate. You can also go to Window->Preferences->Test->Test Generation->HTTP Test Generation and change the option Create new page if delay is greater than 5000ms. You can make that smaller and see if you get more proper pages generated. Your other option is to rerecord but take your time between pages and make sure to count to 5 seconds before performing the next action.
Also remember that when doing recording, we will capture each connection that was used and will send out the requests on the same connection that they were sent on originally. If you had two connections sending requests at record time, we will send requests on two connections at execution time. If you see a really long client delay, then look at that request in detail to determine if it was something that was sent by a user gesture or as a result of the primary request.