To play devil's advocate, that's because the cache of a modern processor is highly optimized for TDM, not SDM. Perhaps a redesigned cache hierarchy could resolve this problem (at the expense of awful TDM performance).