Optimization of language implementations is arguably never premature. When such an optimization translates to an application speedup, it's indisputably not premature.