update2: evdb came up w./ a test harness that works around the problem noted in update1

update1: the 'evdb' destroys the data, so the benchmarks as run are not valid.

<strikethrough>All my previous posts about speed are suspect. My current test numbers have suddenly become inconsistent with my previous numbers, even for relative assessments. I don't get it, but right now it looks like the OP's best is better than my best by a little (and of course the 'evdb' solution leaves us all in the dust).</strikethrough>

--Bob Niederman, http://bob-n.com