I think "performance decoration" is defined as a process in which you make software to look fast but without any positive impact on actual delivered performance (there was an article in ACM a while ago about it).
For example, I saw heavily optimized (and convoluted) C++ loops inside I/O heavy code where just increasing block size and enabling direct IO gives 90% performance boost.
So if there is no "performance lifecycle management" explanation related to optimized code (i.e., which test needs to be run, what is dependency, etc.) then the code is probably just "performance decoration". My theory is: if somebody did investigation to show that this is a valid performance optimization he would probably document all needed tests which led him/her to that conclusion.