To me, the more interesting aspect is not the fastest solutions, but the variance. There's an almost 2-orders-of-magnitude difference between the fastest safe s

To me, the more interesting aspect is not the fastest solutions, but the variance. There's an almost 2-orders-of-magnitude difference between the fastest safe solution and the baseline entry. When looking at an earlier run (before people started copying off solutions from one another), the standard deviation was 10s, which is 3x bigger than the fastest safe solution; even the standard deviation for the top 20 was more than 4s. In other words, clever algorithms and "mechanical sympathy" makes a much bigger difference than unsafe code, and only few of even the top performers got to the point where unsafe code makes a significant difference.