2% is exactly 2%. Whether that is large or small is in the eyes of the beholder.
Recently I commented in this thread or another that the most frequent use-case for power data has been for pacing purposes, especially in non-flat races. If you are willing/able to dig in a bit deeper, though, there are other applications, including looking for fatigue-induced changes in the ratio of power to pace (i.e., running effectiveness) during longer runs. Simply put, some people show changes but others do not, which might provide insight into how to best modify training in the future.
(Related: somebody recently commented on the Stryd FB page that their masters thesis research found a relationship between “form” a.k.a. oscillatory power and running economy.)