So I would guess calculations like these are rather useless in practice, but neverheless, taking the alleged GFLOP number of the A8, 172.8 and multiplying this with a 1.5 core count increase (if the GPU is actually upgraded from the the quad-core GX6450, to this apparent closest candidate, six-core GX6650) would then yield a 259.2 GFLOP number for the A8X GPU here.
Still quite a bit below Tegra K1 (364.8 GFLOPS - though again, I would guess there are other differences to account for?), but it looks like this would push it past the point of having exceeded the old Xenos (240 GFLOPS), for whatever that might be worth.
Putting aside how useful (useless?) this GFLOP counting would actually be in practice, I guess it would still be a far cry from "console-level visuals" if the output of mobile 3D graphics typically favors resolution over scene complexity (and ever-increasing, I guess, with 3x Retina on the way.)