The second form s/x+//g will generally be faster because adjacent x's in the string will be matched in a fast inner loop in the regex bytecode execution code, vs. traversing the whole regex execution call tree for each x.