You make some good points
Feel free to start a wiki page on it in the development section.
One thing that is a big difference between PBRT and lux regarding academic/teaching use is that the PBRT sourcecode is true to the contents of the book,
which is probably why it's got such good educational/academic acceptance.
This is not the case with lux, it's starting to deviate a lot from the PBRT book contents, so I think it will not find easier acceptance for the time being,
especially when it comes to education on fundamentals (like intersection/accerelation etc..., all of which is already in pbrt and the book).
On the other hand regarding the more current/flexible interfaces and more modern algorithms and debugging, your points to use it as an acedemic vehicle
for testing new modern algorithms and other related research makes sense.
Hopefully our first official release will be of sufficient quality and debugged to make these statements correct, which i'm sure it will