"If you mix 2 samples just by adding their bytes together you will overflow (adding 2 8-bit numbers will at worst result in a 9-bit number). So you need to shift right the samples before adding them."
If you have them 8 bits in memory, it's better to shift right AFTER addition (only one shift, better precision). Maybe ggn was speaking about having them pre-shifted in memory, ie 7 bits?