After a thorough investigation I found the problem (and the solution). The EF method (adapted) is compatible with Ecoinvent (elementary flows) except regarding the land use impact category. In this method, there are elementary flows missing (e.g. "Occupation, forest, intensive, normal") that are responsible for the wrong results. After manually adding the missing elementary flows in the EF method that are associated with Ecoinvent datasets, I come up with the same results as in the ILCD method.