The QLN graph is not something that I have ever seen before. 
Jizzerman's QLN graph does not display this kind of characteristic all the time - the one visible on MDWS right now is pretty normal.
However, we have seen a graph a bit like it before, though at a slightly reduced amplitude (mine oscillated by 5dB, while OP's was 10-20dB).
When my line resynced strangely just over a week ago (
discussed on this thread), I got a fairly similar QLN graph too.
That can be seen on MDWS too - if you choose "Quiet Line", then "Historic", then "1st Mar". It was back to normal on the "2nd Mar" after I re-synced manually.
We pondered whether this happened at a time the DSLAM was re-booted (perhaps G.INP going live), so the QLN display looked odd while many modems were re-syncing.
Turned it off overnight. It re-connected back at normal 40/10. Then it REINed about 9.50 and dropped it again to 25/10.
Gave the same crazy QLN during that re-sync.
If it was that severe, perhaps it affected lots of other modems, and they all re-synced at a similar time. If the QLN is a side-effect of that kind of thing, that would imply it isn't just your line!
What would make the DS D3 be 0 for long periods and jump around?
Noise in the frequencies of D3 - enough to make the modem choose to not use any D3 tones during the sync.
I took a look at your SNRM/band graph on MDWS, which showed an event like this happening on the morning of the 7th March.
Then I compared your SNR/tone graph between "Mar 7th 02:00" and "Mar 7th 12:00" ...
... and Wow! Both D2 and D3 look horrible, with dips to 0dB
every 12 tones for long stretches. The dips are 1-2 tones wide in D2, but 4-5 tones wide in D3. The part between tones 360 and 460 look affected too.
The matching bits/tone graph show that the modem has indeed chosen to not use a lot of tones (esp D3).
Yeuch. (Professional assessment, that
)
At a guess, I will say a physical defect in the circuit. Perhaps a defective joint, a joint that can open and close when subject to vibrational disturbances?
Nothing appears to "jump out"; there is no figurative "smoking gun" present. 
I think those dips every 12 tones is a distinct smoking gun - it seems very repetitive, and suspiciously close to 50Hz - and I'd show that to a REIN guy if he ever turned up again. (Grab copies, in case we lose MDWS!)
But I haven't a clue what is causing this. It makes me think of resonance, and fast fourier transforms, and many other things long-ago forgotten.
