It's only cheaper if the underlying memory system has enough bandwidth to handle the throughput of all those cores at once. Otherwise, a fraction of the cores

It's only cheaper if the underlying memory system has enough bandwidth to handle the throughput of all those cores at once. Otherwise, a fraction of the cores end up as dead-weight, which is surprisingly common in x86 systems set up like this.