Benchmarking proved that the second version is about 50% slower than the first & simple one.

Intresting. I know regexes are expensive, but I expect that the second version will scale better. If the number of chars in $word1 is N₁, and the number of chars in $word2 is N₂, then I think the first version would be O(N₁N₂)--pretty bad. The second one should be simply O(N₁), though it depends on how the regex is implemented internaly.