Back in February, Antonio Cangiano did a fairly impressive run at benchmarking the various Ruby interpreter implementations and comparing the results. He's just done it again and, just nine months later, the improvements are fairly dramatic.
I'd summarize here but frankly I hate trying to reduce the complexity of a benchmark suite to a pithy conclusion. Instead, I'll inline one of the several results tables here; this one presents the benchmark execution times in seconds:
I hope he'll provide the raw numbers in some format I can pull into a spreadsheet and play with myself.