(1) One would prefer precision over recall for spam detection, so bear some false negatives; (2) these classifiers are subject to adversarial attacks -- you may see spammers repeating the same tricks to circumvent the detector for a while and providers are slow to update their detectors.