b*cat nods his head after viewing the bit-loading per tone. It does look rather similar to your own.
Again, I would agree that his QLN graph does, overall, look "better" than your own . . . however both of them are better than mine!
Let us look a little closer at the two QLN plots. Notice that the overall level of the first graph is between -120 & -110 dBm, whereas the overall level of the second graph is between -110 & -100 dBm. Now remember that the more negative the overall level (that is the more quieter) the better the graph. So although, by eye, the shape of the second graph looks "better" I would say that the first graph is preferable, as it has a lower average "noise floor".
Moreover, his router shows SNR on unused tones, my router reports 0 dB on unused tones. Is his gap have the same cause as my gap? He lives in other town.
There is really nothing I can say. I just do not know.
Finally, with regards to the two DSLAMs -- all we can say, with certainty, is that they both use Broadcom chipsets. Nothing else is obvious.
I attach, below, my current QLN graph. Notice that I have two distinct "noise floors" -- between -115 & -120 dBm for US and between -130 & -140 dBm for DS. Although the overall "noise floor" is quieter than the two graphs that you have shown, I have severe RFI at various frequencies.