I agree, instead of binning and throwing away who knows how many chips that fail validation, they probably just slapped on a one-size-fits-all (over)voltage, which seems to have backfired tremendously.
I mean, if by undervolting you gain performance while lowering power draw at the same time, RX 480 could've not only looked better in reviews (decisive lead over 970, very close to stock 980), but there likely wouldn't have been an issue with pulling too much power over the PCIe slot. Imagine how much better the launch could've gone then.
As an aside: Intel had the exact same issue with mobile Haswell CPUs. My 4900MQ was a blast furnace at stock voltage, and it wouldn't even run full turbo but 100 MHz less because it was constantly pushing against its TDP limit. I undervolted by 80mV (yes, 80mV), shaved off 7C, and the thing now runs full turbo all the time. But the real kicker for me, was that the stock voltage was enough to get a stable 4GHz OC on my 4900MQ (default turbo = 3.6 GHz). Intel seriously dun goofed there.
It would have leaded to a more expensive product and would be lower in volume. The real solution would have probably been to lower clock targets and add another 6 pin plug (or an 8 pin plug). While it is not typically the case with small chips, it could be that the variance is much larger than expected (they have very sophisticated models for this).
Then again, AMD has traditionally clocked many of their SKUs to the moon at the expense of power consumption, so it has most likely been a design consideration they made. Many of these factors wouldn't have been noticed by an average user. Except for maybe the high current over the PCI-E bus. That could add significant EMI to other PCI-E and integrated devices, although an average user again probably would not be able to pinpoint the GPU as the cause of that.