In the provided example, isn’t the fact that the results change with respect to the software version the real problem?

If the software package used by the author contained bugs leading to type I or type II errors (e.g. by affecting convergence), perhaps it would be best to let “readers” use their own installation and investigate the differences. This could act as a safeguard against systematic errors due to software bugs that would go unnoticed if everyone used the potentially incorrect container.