Ask Questions and Find Answers

This is impossible to answer, there is really zero context here.

You're basically saying that when lots of folks start the same process at the same time, their load time is shorter at the start of the process than at the end of the process.

Okay, but there's no context of what the process is doing. Consider an application for submitting data to file US tax returns. Same kind of deal as what you're describing, it's basically a series of questions where each response goes back to the server. But when we look at what happens server side, US tax firms will build in logic and analytics and active decision making to use each question to trigger other questions, make suggestions, evaluate audit risk, etc.

The farther you are down in this process, the more backend work is going on evaluating the data collected up to the point, therefore the more backend work is involved with each data point, so the exact same kind of load performance could be seen in that kind of process.

That's just one example, but I hope it demonstrates that you shouldn't really be asking why load time increases as the process goes on, instead you should be asking why the process struggles as more data is added. Perhaps it is not caching previous results (or perhaps it can't), perhaps it is doing work that could wait until a final submission, etc.

Exactly the answer I was looking for. I know the question is almost impossible to answer, but since I was stuck in my thinking I needed another look into this. And maybe a miracle saying change this param and you'll be fine. The little context I supplied can be a little more detailed as this scenario:1) Frontend extracts a question from backend and presents it to the user using a web form.2) User answers the question and the frontend send data to backend for storage.3) Backend fetch next question and returns it to frontend and the process starts all over again. One calculation being done that can varys over time is the "Are all questions answered". This methods has been optimized and shouldn't increase over time.But the answer you supplied gave me the simple idé of putting a log print out in the above scenario and examine how this change over time. With time simple test I can eliminate or pinpoint if the problem relates to code or environment. Should have thought of that earlier. Thanks.

Sometime a fault can lead to discovering things. In this case it turned out that the test tool we are using was loosing a vital settings which in return made the scenario never reach backend. The scenario instead became:- Login to site
- Start inquiry - For every inquiry, at the start of the portelet rise an error. The error was displayed on an error page using an error portlet.
And then the next question was submitted from the test tool. New error, error page and next question. Repeated for all questions.
So the test never reached the back end. Major part of test produced two pages:
- question page
- error page

I still can't explain why page load time grows from 2s to 4s with similar load. Any suggestions on where to look?

Tomcat hast a concurrent connection limit of 150. Maybe you hit that? You should see a warning about it in Catalina.out.
Or just try to increase it in server.xml and test again

Thanks Christoph! I'll will look into this. During this weekend I did some more load tests and due to other problems I had to descrease number of virtual users down to 100. Is Tomcats concurrent connections based on one user one connection or can one user have more than one connection?Since I am only using 100 virtual users I shouldn't reach this limit but the graph of page load time grows the same way as with 150 VU.