I compared the DVD AC-3 to the US LD AC-3 a while back when borisanddoris sent me his bit-perfect capture. At the time I noticed that the two releases had completely different LFE tracks but never went any further with more analysis. FoLLgoTT's post got me curious so I had another look at the Live DVD AC-3. It really does seem that the LFE/subwoofer channel is stored in the surrounds.
The only explanation that makes sense to me is that the 'CDS' mix Live used was in fact the 70mm 6-track master and it was handled incorrectly. T2 was presented in 70mm Dolby Format 43, which is baby boom with split surround. 70mm only had the one dedicated surround track so stereo surround was stored in tracks 2 and 4, in the unused bandwidth of the baby boom (subwoofer) tracks (much like theatrical DTS). The mono surround track was a sum of the stereo surrounds for backwards compatibility with Format 42. On playback the baby boom tracks are high/low pass filtered to separate the surrounds and subwoofer, and the mono surround track is filtered to add back the missing low frequencies to the stereo surrounds.
Looking at the Live 'CDS' AC-3 track it looks like the surrounds were transferred as-is, keeping the subwoofer information, while the mono surround track was transferred as the subwoofer track, when played back on a monitor 5.1 system it would have delivered a rumble (and the higher frequencies would have been filtered out by the subwoofer's low-pass circuitry). Upon encoding to AC-3 it would have been low-pass filtered automatically. So potentially this is the untouched 70mm theatrical audio, just presented incorrectly. The US/JPN AC-3 laserdiscs which were released a year later have correct LFE channels and may be the CDS mix, but could also be the 5.1 mix used for the Special Edition laserdisc (which may or may not have been tweaked).
Low-passing the surrounds in much the same manner as a DTS APT-X100 file results in a LFE whose waveform is very similar to the US LD AC-3. After high-passing the surrounds I think the low frequencies can be restored with the information in the LFE channel. This may be an interesting experiment, as the DVD AC-3 might be slightly more dynamic than the LD.