Grab,
What happens with JT files is that they are exported and can be maintained up to date as such from NX for use with separate viewing software. JT2GO, Vis-View or Teamcenter Visualization, all versions of much the same thing with various levels of sophistication and they go by a few different names I think in slightly different guises so that I'm never quite sure what to call them. Anyhow they create these files which become can work as assemblies. So far good point it has possibilities.
You can open both file and assemblies in NX-5 and indeed in NX-4 from output made with NX-5. However there are some serious limitations firstly if it is an assembly that you're opening then the assembly structure is re-created with kind of a double level for each part having two items in the ant. These can be used and saved as part files if needs be as I'll explain. If you open NX-5 output in NX-4 then the names come out with appended a few extra characters that also don't appear in the original and so aren't quite right. So as far as exporting and importing the assembly structure JT is slightly inferior to STEP214.
Now as regards the actual geometrical data, the files are smaller than Parasolid exports, which is slightly odd since they eventually contain approximately the same thing. I say eventually because the JT data initially imports a faceted version of the models, but you can go into the file properties and extract the geometry to what then results in a unparameterized bodies. The data once extracted seems to be every bit as good as a Parasolid file. While I have seen curve sections exported into JT files used in teamcenter I've not yet discovered how it can be done or whether it can help you get those curves via the JT file back into a previous version of NX. Certainly it is surprising that the data can be extracted so successfully from a smaller file, but if working in an assembly each component has to be done individually, and the extraction is a fairly slow process.
So there you have it not a bad idea, but no parameters certainly no drawings, and few if any entities in fact other than the solids. But better quality data than STEP214, and something of an assembly which if not quite right is certainly not wrong at least.
Because it is a separate file there are still parallels with any other translator, but it is certainly a method I would consider using and perhaps if some of my gaps about the whys and wherefores of curves and other entities can be addressed or responded to here then it has potential. It has even more potential is we can find ways to use it or leverage it into some sort of method that enhances the product towards having the type of backwards (lets not say compatibility as that implies too much), call it capacity to open newer files, then that would save some users a certain amount of pain in the future.
Best regards
Hudson