Actually, my personal opinion is that you should always use the one you find the most comprehensible. In this case, I find grep the easier to understand because it's being used as a filter. In a more complicated real world example, a loop might make more sense, especially if the filtering requires a lot more than a regular expression.

It's not time to worry about which one is faster until the one you're using is too slow. Profile your code first and then look for speed improvements in the places that matter.

Figuring out the fastest way to do something for its own sake has entertainment value, but it shouldn't be how you decide to write something. Write for those who will read it (and by this, I do not mean the interpreter).

Besides all that, in this case, I think testing shows there isn't a significant speed difference. The biggest difference between identical data sets was in the 100 case, and even there the difference was only 22%.