My conjecture, and I readily admit that it's just conjecture, is that TX1 is a close match for the performance of the final hardware, that also happens to be architecturally similar to the new custom Tegra in the NS. I'm assuming that if the more aggressive speculation about NS was correct, earlier dev kits would have used a pascal based board to better simulate the performance of final hardware.
It's worth noting that your example of early Orbis dev kits is more similar to my hypothetical pascal based board. Early Orbis dev kits were octo core bulldozers connected to a discrete GCN GPU, which gave a good approximation of PS4's GPU performance over other specifics like bus level architecture or even CPU architecture.
Also the timeline for PS4 dev kits had units with near final SoCs delivered in January 2013, ~10 months before launch. NS dev kits based on TX1 hardware were being used as recently as July. Given a March 2017 release date for NS, that means you would have to believe that Nintendo, or certain third parties at the very least, were still relying on very early, crude approximations of final hardware only 8 months before launch.
Now maybe I'm just a cynic when it comes to Nintendo speculation threads, but absent any well publicized and heavily vetted leaks that would give me reason to reconsider, I think I'll continue to be suspicious of any and all claims of near home console performance in a handheld.