"The tests ran on a desk(t)op with 16GB RAM" a single node setup doesn't fit the distributed, fault tolerant, use case that Spark was designed for, rending this comparison moot.