[noxin]

They likely benchmarked it on a 32 core system. Like a dual Opteron board. If the task was single-threaded before a 32-fold improvement is reasonable.

[pedrocr]

It's very difficult to get a 32x speedup from 32 cores as there are always parts that are inherently serial, so it's more likely they tested it on a 64 core machine or something like that.

[jelder]

Yes, this is thanks to Amdahl's Law.

https://en.wikipedia.org/wiki/Amdahl%27s_law