See this report: WebPageTest - Running web page performance and optimization tests...
The first run is great. Love it. The second run is a disaster. This result is reliable over many months of samples. In fact, first-run time is trending down over the last few weeks while second-run time is trending up (by a LOT).
I don’t understand why that might be. Any idea what would make the cached load take so much longer to get off the server?