(You did not say how many simultaneous users you need to simulate for your stress test. My answer assumes you need to simulate more than 5 or 10.)
Using your Selenium tests as the basis for stress testing is an attractive idea. After all, they are already written/debugged. You would like to check the same things under load, so they seem relevant. And you would rather not have two different kinds of tests, using different tools/frameworks, to maintain. If one instance of a Selenium test is a functional test, couldn't one thousand instances be a stress test?
Unfortunately, there are some issues that get in the way of reusing your Selenium tests for stress testing; see for example Looking for open source, Java-based tool for performance testing. Here are two issues:
- It is expensive. You cannot run many instances of Selenium on the same machine at the same time. In my experience, you need at least ten times as many machines using Selenium as you would with JMeter.
- Much of the logic in your function tests is wasted on a load test. The kinds of bugs you will find running at scale will be different from what you will find in a functional test, e.g. lock contention and algorithms that do not scale well.