I don't remember 50% gap in resolution/performance on PS4 and X1 being common (very few games were 720p vs 1080p, most would be 900 vs 1080 or 720 vs 900), and don't think that was due to the edram.
So yes empirically you're absolutely right most were not the full 720p vs. 1080p. I didn't intend it to be read like that was the norm, only that in extremis you get a 50% resolution hit.
The reason I say it has to do with the RAM (and by extension the ESram) is because of the way it was partitioned up. I'm not a developer obviously but to my limited understanding, unless you can fit your render target into the 32MB buffer of extra-speedy RAM, you're forced to relegate it into the DDR3. This meant that, especially in multiplats which seldom used the buffer, the Xbox One would construct its frame with DDR3 (which is supposed to be system RAM) while the PS4 was able to use its GDDR5. The bandwidth differences then become seriously 2.0. The new Series S has a similar tiered memory architecture which people suspect the slower 2GB will be used for the OS. But even the faster memory (8GB GDDR6) has less bandwidth than the slowest tier memory in the Series X. So to my mind, unless developers scale their rendering targets nicely, both for the GPU but ALSO for the memory bandwidth, the Series S will get the short end of the stick. In practice that means that cutting texture resolutions AS WELL as internal resolution is basically non-negotiable. Watch Dogs got it right and AC Valhalla got it wrong