Substantial improvements to coercion optimisation