Again, I point you back to understanding what bearer 0 and bearer 1 are for.
Bearer 1 does have a say in latency
Bearer 1
INP: 4.00 4.00
INPRein: 4.00 4.00
delay: 3 0
PER: 16.06 16.06
OR: 95.62 31.87
AgR: 95.62 31.87
Look at the bottom line of this: The "AgR" line - or aggregate rate.
Bearer 1 carries 95Kbps. Compare that to bearer 0, which carries 80,614Kbps (at least it does on my 80/20 line).
Bearer 0
INP: 46.00 47.00
INPRein: 0.00 0.00
delay: 0 0
PER: 0.00 0.00
OR: 0.01 0.01
AgR: 80614.82 20102.08
Which do you think is carrying the main stream of data?
Another clue comes from the OR data (Overhead Rate). It seems that bearer 0 has almost no overhead, but bearer 1 is entirely overhead.
The final giveaway comes from reading G.998.4 - the G.INP specification - section 9, combined with the annex that applies to VDSL2. The spec uses different numbering, but it puts the user data into bearer 0, protected by retransmission. And it puts the "overhead channel" (which sends management data back and forth) into bearer 1, protected by FEC and interleaving.
Result: 0ms delay to user data. 3ms to the overhead channel, sending management data back and forth between the modems.
Read adslmax post as he had delay set to 3 and it was removed for some reason and is now set at 0 and his ping did drop by 3ms backed up by thinkBB graph.
With the best will in the world, I wouldn't put Max down as the most stringent observer of experimental behaviour.
Max and I have one thing in common... we both use Plusnet. It is observed behaviour for Plusnet connections to change gateway at each resync. It is also observed behaviour for latency to alter within 1-2ms on different gateways, and I suspect that one gateway is even worse.
With a natural tendency to not want to switch too often, it makes it hard to confirm whether small latency changes come from landing on a slower/faster gateway, or from a change in modem settings. Distinguishing one from the other takes time, and careful observation. No jumping to conclusions.
So lets swap to an experimental investigator I trust a little more. One who makes more careful observations, and is less prone to rush to judgement...
So far, I have seen a max of 1ms of variation in my end-end ping times since G.INP activation, having swapped through 3 different gateways. Attached is my BQM of the day G.INP activated (snipped, to fit to the Kitz size requirement); you see a tiny increase when G.INP activates at 06:30; a slight decrease after a resync at 19:15; a slight increase at 22:15. All 3 resyncs resulted in the same configuration of both bearer 0 and bearer 1 as I mention earlier.
Yet I had the same exact configuration - 3ms "delay" in bearer 1, only downstream, alongside INP of 4 and INPrein of 4 in both directions - introduced at 06:30, and left in place at each of the other resyncs.
So what scale are those changes? Zoomed in, I can see that 20ms on the scale is represented by 28 pixels - so a pixel is approximately 0.7ms. The visible changes in the graph happen at the scale of 1 pixel, less than 1ms.
You certainly don't see 3ms extra latency switch into effect in any of my test results - either this BQM, or the previous graphs. What you see are the kind of changes I'd expect by swapping gateways at Plusnet.