To be honest, I couldn´t completely understand EasyFX explanation, but it seems that he is using some out-of-sample data, so good enough IMO.
About your answer, I don´t know how do you find a "lasting parameter", but if by "finding a lasting parameter" you mean not using OOS, sorry but I do no agree with you, you will be likely overfitting. Another important aspect when backtesting an strategy is that the in-sample period should be large enough.
The article you posted explains a situation where a WF doesn´t work and a type of Cross-Validation works, I could say to the author that the opposite situation could be found...
He also explains what in the end is a Cross Validation approach; this can be done easily with the Sklearn function Cross_validation_predict, which is another way to backtest a strategy.
If you think carefully about the WF approach, you will notice that in order to train the model, you could use Cross_validation to build the model within the in-sample-data, so you will cross-validate in-sample and WF out-of-sample.
In any case, WF is one tool to backtest an strategy. I hadn´t seen it implemented in any standard python machine learning libraries, that´s why I posted it. At the end, everyone should use the backtest method that best suits him/her.
Have a nice day.