So, two things here.
The morphing of old drawings to new data could have been automated if the process of sub-selecting from the whole survex dataset could be formalized. Only by survex processing the whole cave system and then snipping out the cave section you want, instead of compiling from some sub-directory, can you get the legs into their real world positions with the final loop closures applied.
Secondly, the bezier lines output by tunnelx will be accurate and should work for Therion. If the segments are too short and annoying, I could write a script that takes in a valid therion file, combines them into longer pieces and outputs them as longer combined splines. This script could at least be used for other purposes and would be easier to debug as you wouldn't need TunnelX to preview the input and output.
But the main issue is -- as Mad Phil has pointed out -- by the time we do all this and pester one and a half people to train themselves and be interested in ploughing through all this work to eventually produce an output after a number of years that's pretty much the same as what you have already, the era of cheap lidar scans and handy intuitive 3D representations will be upon us. So this is a bit like investing time into a new revised edition of the London A-Z Streetmap booklet just as google maps and OSMand are coming out on these new-fangled smartphone thingies that are never going to catch on.
Julian.