The work of Pedersen et al. (2009) is one of the examples of
using TS as the base of solution algorithm. They use the instance
sets of Crainic, Gendreau, and Farvolden (2000) and Ghamlouche,
Crainic, and Gendreau (2003). In comparison to a MIP solver, their
algorithm shows robustness, and even outperforms the solver in 33
cases of the 78 instances they tested. Although the tuning of TS is
dependent to the instance characteristics, in instances with particular
structures like high fixed-variable cost ratio and/or loose
capacity, TS outperforms the MIP solver. With the use of an independent
multisearch strategy, Pedersen et al. (2009) could furthermore
improve the results to 2% more. However, their TS still stands
as a powerful algorithm.