Importance sampling is a technique that can significantly reduce computer run-time in the estimation of bit error rate (BER). However, in the conventional implementation (CIS), the improvement reduces markedly for systems with long memory. An approach to recover the full improvement for such systems has been previously suggested, and is called 'efficient' important sampling (EIS). A report is presented on an extensive series of simulation-based experiments with CIS and EIS, both to compare theoretical predictions to experimental observations, as well as to gain insight into the conditions of applicability, especially for EIS.>