However, the DLM applying banding without INP is akin to increasing the SNRM, still without INP. So I suppose it depends on what the DLM is trying to protect against. If the DLM is trying to increase resilience against sources of constant noise, then it might increase the target SNRM, or cap the rate, which both achieve the same result.
Simplistically, it is equivalent.
However, I think banding is (in a broader sense) considered to be a newer, superior DLM mechanism than increasing the target SNRM, and is described as such in NICC's document on DLM. This might relate back to power settings, and might not fully apply to BT's model.
Alternatively, if it's trying to protect against pulsed noise sources, either isolated pulses, or repetitive (REIN), then it could set INP. Also, it wouldn't surprise me if BT try to avoid any form of INP to avoid the equipment using slightly more electricity from doing all that extra processing.
If BT were that worried, surely they wouldn't have bothered with G.INP, where all that extra processing is turned on anyway?