I was actually about to make a post about this same topic, when I noticed this thread that Arch started.
At Ask Mr. Robot we have been recommending high Haste for fury since Legion launch, since we saw the same trend in our simulation results as Arch. We were getting dozens of questions about the 18% supposed haste cap for Fury, and always telling them to just gear for Haste. Glad to see some consensus on that approach is being reached. (And that I'm not crazy.)
One thing that I have noticed is that Crit and Mastery are really falling within margin of error of each other in the simulations I have done. As long as you are favoring Haste first, picking mastery or crit as the second stat just doesn't seem to matter that much.
I've improved on the Fury rotation in the AMR simulator and noticed a few things in the SimC APL that could be improved to get better results with some of the other talent choices:
1.) Massacre and Frothing Berserker are really just about equal in value in my tests. The SimC APL is allowing Rampage to be used at less than 100 rage when you don't have enrage, which is reducing Frothing Berserker uptime over the course of the simulation, devaluing it slightly and making Massacre look a tad better compared to it.
2.) Inner Rage is still the only real option for single target damage, but, to get a fair comparison we need to do good simulations of the other two talents.
2a.) The SimC APL needs to make sure to refresh Frenzy if it is about to fall off. It's letting Frenzy drop and making the talent look even weaker than it already is.
2b.) SimC is delaying Bloodbath until Battle Cry is ready and this results in less Bloodbath uptime and lower damage overall. Delaying it for AoE is a DPS gain, but, on a single target I'm getting much better results by using it on cooldown instead of waiting for Battle Cry.
Here is a comparison of talent choices that I did, just for curiosity. I don't think there is anything unexpected here:
http://www.askmrrobot.com/wow/simula...053fdeed68640a