Performance questions are not answered easily - especially without code.
Why don't you just try both variants?
Then run the test 10 (or even 100 or more) times and get the average value.
This should even out the side effects for both methods - to compare good enought...
If they are "to close together" for a definite answer you should use the "safe way". Because IMO unsafe
should only be used if really necessary.
P.S.: I can't recommend any books or similar to such methods - "learning by doing" is my advice here. In such cases I also just try different things and analyze them, maybe search for different approaches or ask here for specific help with a short code example (already done here multiple times successfully). After that I most of the time have better and more performant code and learned something new...