Rather than doing the full-blown teddy implementation, it could potentially have a significantly tighter inner loop by doing what we do today for IndexOf("...", OrdinalIgnoreCase), e.g. creating a vector for two characters in each string and comparing those against the input at appropriate offsets. This will require prototyping and measurements to see if it's actually a win; it's an idea @MihaZupan and I discussed as a way to possibly mitigate overheads when searching for strings that have starting characters that don't match very frequently, in which case IndexOfAny(startingCharacters) does fairly well.