There is another pure Java implementation that could be included in the benchmark, its called JSnappy. A nice feature is a optional multi thread decompressor. It would be intresting if it is faster than -> "https://github.com/dain/snappy".
Interesting. Wonder why they felt the need to reinvent the wheel, instead of collaborating (wrt core decompressor).
But yeah could be faster for long enough data: should be easily parallelizable on per-block basis.
Even there license is the same: "Apache 2.0"...