We can do better than "every C program is a valid C++ program, so C++ can’t lose" defensiveness.

Look at the data tables published with that 2017 paper:

https://sites.google.com/view/energy-efficiency-languages/re...

There's an order of magnitude difference between the times of the selected C and C++ programs, for one thing — regex-redux. That is enough to explain all of the difference between the C and C++ time measurements shown in "Table 4. Normalized global results". (Seems like an outlier which could have been excluded.)