Yes, because it will extract literals from your regex and seaech for those first. When a match of the literal is found, all you then need to do is find the line boundaries of that match and run the regex engine on just that line. It ends up working pretty well.

Of course, if your regex has no required literals at all, then this optimization can't be performed. GNU grep can slow down quite a bit in these cases, especially if you have a non-C locale enabled and use some Unicode aware features such as \w.