Video traffic now represents a growing proportion of the traffic on cellular networks, causing capacity problems for operators and increased delays for users. Studies have shown that deploying caches at the network level reduces the delay for the end-user and the overall traffic volume for the telecom operator. In this paper, we analyse a large nationwide dataset of real-life video requests sent by mobile users to a popular video streaming website. This analysis is the first to rely on such a large dataset, and sheds light on the optimal cacheability of video content with caches distributed in the cellular network, and how efficient some existing cache replacement algorithms are at reducing the number of requests sent to the video provider. We show that depending on the cache size and algorithm parameters, up to 20.33% of the requests can be served by a local cache.