I agree with most of the comments here that 44% faster sounds way too good to be true. Perhaps there is some micro-benchmark where this gain was perceived. A more through blog post to substantiate or disprove these performance gains would be welcome.
It'd be really interesting to see the methodology used when gains like this are reported, because it's potentially really interesting stuff but without knowing what was measured and changed, it's hard to give it weight.<p>Which is a shame, because it's probably really useful.<p>i.e. what's the workload. what was the performance metric which improved. any detail you can give about the likely bottleneck of the system, etc.