It's very plausible the Vega56 will consistently beat the 1070, if the Vega64 consistently matches the 1080.
The 1070 is actually very cut-down. Being cut-down 33%, or having 0.75x the shaders of the 1080. As well as lower memory bandwidth, and a worse form of memory so there's no chance to overclock to the same as the 1080.
On average it's ~25% slower than the 1080, and this can grow to 30%+ in highly optimised games like DOOM: https://www.techpowerup.com/reviews/Zotac/GeForce_GTX_1080_Mini/28.html
The Vega56 is only 14% cut-down, or has 0.875x the shaders of the Vega64. If it can clock to the same level, and there's no reason to believe it can't, it will be less of a drop compared to the Vega64 as the 1070 is to the 1080.
Therefore the only way the Vega56 won't beat the 1070 on average is if it clocks terribly for some reason, or the Vega64 consistently loses to the 1080 by ~10%. And from what we know at the moment, neither of those seem to likely.
Also I say this as a 1070 owner.
If you ignore cherry picked games and look at a broad average, the 1080 is 20% faster than the 1070. No need to use theoretical numbers based on cores, we can use empirical measurements:
https://www.techpowerup.com/reviews/MSI/GTX_1070_Quick_Silver_OC/29.html
E.g., a vanilla 1070 is 95% of an overclocked 1070, and the 1080 is at 115%. 115/95 ~=21%.
And of course the 1080 is only 15% faster than the base 1070.
Given Vega clocks and core count I expect Vega 56 to be very close to 1070 speeds, but possibly slightly faster. This then makes the Vega64 look pretty useless because you pay a lot more to get 16-18% performance but ti pulls a massive amount more power. If it was possible to clock the 56 to 64 level eaislly then many AIB cards would be at the heals of the 64, which makes the 64 very unattractive.
I do beleive the 56 is probably the most attractive of the 2, but if you are a die hard AMD user coming from a FuryX it will not be an upgrade worth the money, you would have to stump up for the 64