Current performance tests are greatly affected by JVM warmup (e.g. Java's tiered compilation model). A few options:

• Implement JVM warmup in our tests by executing each test operations many times, and only recording last result (needs to be done on a per-test basis)
• Use Spock's EmbeddedSpecRunner to run entire existing test suite many times to warm up the JVM (only record last result)
• Use Java Microbenchmark Harness (JMH) framework, which is annotation-based and can be easily configured to do separate warmup and measurement iterations. There are many other benefits of JMH, so this would be preferred solution, time permitting