[tcoppi]

It could even help with low threadcount workloads since the L3 will presumably be able to be fewer cycles away from each core than a unified last level cache would be.