Lexical effect handlers offer both expressivity and strong reasoning principles, yet their practical adoption has been hindered by a lack of efficient implementations. As effect handlers gain mainstream traction, it becomes urgent to show that lexically scoped handlers can be made performant. My doctoral research addresses this need through the design and implementation of Lexa, a language and compiler that achieves state-of-the-art performance with lexically scoped handlers. Drawing on deep semantic insight, I developed techniques that advance both implementation and theory. Lexa outperforms existing compilers, and all techniques are formally proven correct.