JLP's kangaroo for example is definitely not fully optimized when it comes to this.
Check out Etarkangaroo if you haven't, it seems to have some improvements, albeit small.
Max range width up to 192 bits and (on my setup at least) it's 1Gk/s faster than JLP's.
I have no idea if it's fully optimized but I'm guessing it could be a better starting point for further improvements, although one potential downside is that it requires PureBasic to compile, which has to be bought.