You're right that we don't use Spark as our data engine, but we also don't use hyper yet either (well, we do in beta[0] :). Hyper is replacing our existing data engine (conveniently named "Data Engine" [1])
Yeah you're right about spark for the processing. I did see a lot of Hadoop binaries in their tableau server (it amused me to see zookeeper.exe in windows task manager) and assumed they were using spark for internal data representation, but that was a false assumption I suppose.