I’m personally somewhat iffy about VDFs as well 
The dependency on a delicate balance between commodity ASICs and attackers doesn’t really sit well with the “be the scalable blockchain that can survive WW3” desideratum.
Though the nice thing about all of these approaches is, I’m pretty sure we can stack most of them together and get the benefits of all together. I know that’s true for plain RANDAO + VDF (+ my GHOST-related work in making the algo more robust to bad randomness in the first place); I imagine it could be true for low-influence functions as well.