In terms of speed and efficiency which is the "best" supercomputer, NVidia Tesla, CPU systems (ie. Cray XT5, IBM's Roadrunner) or a distributed computing type scenario?
TMK, nobody has posted Tesla frame times or approximate PPD figures, although some have speculated nearly identical production values to a GTX280. It would be interesting to actually get some hard numbers, though.
@Phantum, it depends on the type of 'work' or application you want to run. DC can be very productive due largely to untapped potential but where it shines is in projects that require many, many years of continuous research. F@H is a good example where DC is put to very good use. Other scenarios such as meteorological modeling or other types of modeling may benefit more from dedicated specialized clusters. GPU-based computing is relatively new and might one day supersede 'traditional' supercomputer designs such as the ones you mentioned. Look where it has taken folding in only a few months.