State-dependent importance sampling (SDIS) has proved to be particularly useful in simulation (specially in rare event analysis of stochastic systems). One approach for designing SDIS is to mimic the zero-variance change-of-measure by using a likelihood ratio that is proportional to an asymptotic approximation that may be available for the problem at hand. Using such approximation posses the problem of computing the corresponding normalizing constants at each step. In this paper, we propose the use of path-sampling, which allows to estimate such normalizing constants in terms of one dimensional integrals. We apply path-sampling to the tail of the delay in a G/G/1 queue with regularly varying input and argue such procedure, combined with a change-of-measure proposed by Blanchet and Glynn (2007a), yields an estimator with good relative accuracy with quadratic computational complexity (as a function of the tail parameter).