This might be of interest:
http://www.ihpva.org/HParchive/PDF/hp50-2000.pdf
The first article tests the efficiency of a derailleur drive-train.
They conclude that efficiency varied between ~81% and 99% and that the main factors were sprocket size (big-big is more efficient than small-small) and chain tension (high tension more efficient than low tension).
They say that the impact of lubrication and cross chaining is negligible. However, their test shows an efficiency difference of ~2% between different lubricants. This is in line with the OP's article which claims a 5W difference at 250W input (2%); not enough for most of us to care about but possibly significant to a pro.
They may be claiming that it is insignificant due to probable systematic error in the experiment; I imagine an electric motor in a lab might not be the best simulation of legs in the real world.