How much of a performance hit would they really take if they included word boundary metacharacters in their filtering regexes? Is this something people have benchmarked? Can anyone make a reasonable guess?

(I'm honestly curious, and wondering if this isn't an engineering trade-off as opposed to just a dumb mistake.)