While you might not want to do this with actual Twitter, any sort of high-performance computing workload can run substantially faster on a single optimized mach

While you might not want to do this with actual Twitter, any sort of high-performance computing workload can run substantially faster on a single optimized machine than on a distributed computing environment. I learned this the hard way when I was running a medium-sized MapReduce job in grad school that was over 100x faster when run as a local direct computation with some numerical optimizations.