Couldn't performance variation on a shared server be due to "noisy neighbors?"

(Edit: Assuming that the article doesn't leave out details about how tests were performed, these numbers feel "single data point"-ish to me.)