Inferno is definitely fast, but these benchmarks are highly misleading.
We're working on improving them so they better reflect the difference in keyed vs non-keyed implementations [1][2]. When ready, Round 5 should be a much more realistic apples-to-apples comparison.
Inferno is gonna be really tough to beat, at least for non-keyed implementations. We don't have a keyed implementation from them, so it's hard to say exactly how they do there, but i dont doubt that it'll be at or near the top. For now kivi is likely the fastest possible but the way you have to write your code with kivi is rather unpleasent. However, the whole point of kivi was to push the perf boundaries, not to be useable directly.
At the end of the day, there are numerous <= 15KB libs that come within 3%-15% of Inferno, so the perf and size difference is really becoming secondary to lib ergonomics and features.