I think the long stagnation on 28nm may have biased some expectations about what kind of average yields should be obtained. Nvidia doesn't seem to be doing particularly stellar on yields either, with even the ultra-specialized P100 being a cut-down chip. And the GP104 in the GTX 1070 is cut down by a full 25% of shaders, more than expected. While that latter decision could be the result of market segmentation, I think it has to do with the rush to production and the fact that TSMC is still a relative newcomer to 16FF+.
We'll probably be seeing mediocre yields for 6 months or so. Maybe up to a year. For reference, it was in October 2013 that AMD released the first wave of R9 200 series rebrands, which (temporarily) did away with cut chips for Pitcairn and Tahiti. That was about 22 months after the first 28nm GPU (7970) hit the review sites.
Samsung's FinFET process is known to be denser than TSMC's. The Apple A9 smartphone chip was dual-sourced on TSMC and Samsung, and the Samsung version was about 9% smaller. And that was with the Samsung 14LPE process. 14LPP (the version GloFo adapted) is supposed to provide even smaller size, as well as better performance. Also, don't forget that even on the same process node, there is a trade-off between density and speed. Nvidia could have crammed more shaders in GP104 in the same die size if they wanted, but then they wouldn't be able to hit the 2000+ MHz clock speeds that has everyone so impressed. Since at least Kepler, Nvidia has preferred to go for less dense and higher clocking designs, while AMD's GCN products are generally oriented around denser designs with lower clock rate caps.
In addition, we've all been assuming that Polaris 10 is a 232mm^2 chip based on some information written in an old LinkedIn profile. This belief was bolstered by AMD's positioning of P10 as a "mainstream" part, the modest TDPs, and the low pricing announced at Computex. But it's at least possible that we could be wrong, that the 232mm^2 GPU was something other than Polaris 10 (a future product, an engineering sample that never made it into production, whatever else) and that Polaris 10 is actually closer to 300mm^2. Only time will tell. Fortunately, we don't have all that much longer to wait.