It's certainly not perfect but neither is a goals stat when it comes to predicting future outcomes. When a goalie let's in a stinker, that's not representative of offence created by the opposition, but these sceneraios are generally going to balance out, so while I can point to specific scenarios where either stat fails, you need to look at the forest, not the trees if you want to get a more reliable sense of what the future entails.
The one thing xG stats do well is remove goaltending from the equation, and besides, if your team is wiffing on passes so frequently that it's affect the metric, well, that will probably show up in goals stats too. The problem they have is they also remove finishing ability, so a muffin of a shot is weighed equally to a sniper wiring it. Tkachuk is a bit of a poster boy for creating a ton of xGF without actually creating high quality chances.
I'm not going to argue xG are the be all and end all of stats by any means, but I definitely feel they add something to just looking at Goals stats, the more data the better, with goals you only get 200-300 data points for a given team, with xG your looking at 3000-4000. There will be some garbage in either data set, but I guess the hope is the size of the sample is enough to smooth out the garbage. I think the smaller the games set you have, the better off you are looking at xGF. Where it evens out I'm not sure.