We place in the context of Information-Centric Networks a special class of segment-based caching, named partition-based caching, whereby information items are cached in partitions at different caches in the network. Our rationale lies in exploiting the merits of parallel data transmission, which however may come at the expense of increased network utilization. The study of this cost-benefit trade-off underlines our work. Following a theoretical analysis on the benefits of partition-based caching over single replication caching, we propose three traffic engineering algorithms for assigning content partitions to caches and respective routes. Simulation results indicate that partitionbased caching can indeed improve average content transfer time (up to 65%) at the expense of a tolerable increase in network load (up to 25%), for moderate numbers of content partitions (less than 6), balancing at the same time network load better (at the level of 50%), compared to single replication caching.