Lower cache per core is actually a pretty natural outcome with the latest device fabrication nodes shrinking logic while leaving the size of SRAM largely unchanged. We may perhaps also see eDRAM (a lot denser than SRAM) for last-level caches.