Hi,
The following is my current understanding. Note, not all bullets were verified in a real-world life, might appear to be incorrect (at some points) and I will appreciate comments / corrections from others.
-- Initial assumption is that user actions (read - scenarios) do not depend on on the previous actions of this user. For example, if the user of the tested automotive trading web application looked at Ferrari, it is out of our scope to deal somehow with the fact that he/she at the application's sidebar will be suggested also to take a look at brand new Lamborgini as well. Things like that should be verified via functional tests and in our scenario we will just know that one or several requests for the sidebar are executed along with the main request(s);
-- Recorded scenarios must not be long and complex. The reason for this is that as long as the given web page evolves, the number of requests that it generates as well as the target of these requests may be changed by developers. So it should be easy for you to re-record the changed traffic;
-- You may craft more complex scenarios by creating a new blank scenario and populating it with the already existing scenarios using the Call Scenario operation;
-- With all required scenarios ready, consider their distribution in a real world using logs from your web server. E.g. 60% of visitors just browse for one car, 35% browse for two cars; 2% additionally look for the list of distributors, 0.2% do a purchase, etc. Based on this distribution, consider the number of virtual user groups, connection speed and browser for each group, the number of users in each group (http://blog.smartbear.com/performance/how-to-determine-your-virtual-user-quota/, https://smartbear.com/learn/performance-testing/load-testing-vu-calculator/ and https://www.google.com/?q=load+testing+required+number+of+virtual+users), what scenarios will be executed by this or that group and whether or not the test is worth to be distributed on several machines;
-- Create Tests using the results of the above considerations.