Using a test environment that is as close to production (or as live-like) as possible is a good step toward ensuring reliable results. You can try and use a smaller set of servers and then scale your results up to get an estimate of live performance. However, this assumes that you have an intimate knowledge of how your application scales and what hardware constraints will be the bottlenecks.
A better option is to use your live environment or, rather, what will become your production stack. You first create a staging environment that is identical to live, then you deploy your code to it, and then you run your full test suite, including a comprehensive performance test, ensuring that it behaves correctly. Once you are ...
