...downsampling was made during the copy from rendered tile (eDRAM) to frame buffer. (usually SDRAM)
That's actually the question here; irrelevant of the method employed I can't figure out why eDRAM would be required at any price. Especially if it resembles to the description above. Unless of course Acceleon doesn't have a tiled or chunk based memory subsystem at all, which doesn't make sense to me either.
...but basic idea was adaptive supersampling
Even if then why even bother with a post blur filter if the above should be correct? I've grown highly reluctant lately with any fancy marketing descriptions and no the bell obviously doesn't ring for BB OY at all here.
In any case hypothetically adaptive in what sense? Coverage mask based techniques alike?
I'd say that, it is not impossible, but there's right now nothing indicating it either.
There's no real indication either that NOKIA will even continue with the nGage project, after the relative flop of the initial offering. I'd say chances are even higher that they might even abandon the idea of a real 3D capable nGage2 all along.