Remember the benchmark tests I linked a few days ago?  There’s been ongoing discussion about them, and a recent comment from Matthew Dillon sums it up pretty well: the benchmarks differ depending on whether you favor reads, or favor writes.