It took me a moment to ponder the effectiveness of mapping utf8 characters into a bitmask. At first, it seemed unwise. Then I realized that 32 bits would get you `a-z` plus six special characters, and 64 bits would get you 'A-Z' (uppercase) plus six more. That's plenty of room for HTTP headers, and for a blazing fast matching algorithm since it just masks and compares a couple integers. Even figuring out which bit corresponds to which character is a simple lookup into a 256-word table.
One thing the author leaves out is this technique is technically a Bloom Filter. I find these kinds of things fascinating since they came about in an era where computing was far more resource bound than today (1970 in this case). But here we are, still finding real-world corners to use the same old optimizations.