This paper shows that managing the reliability of knowledge built collaboratively by an e-community improves the results of a reasoning process. Our approach for building a reasoning system over knowledge coming from an e-community addresses first the management of the reliability of knowledge units provided by non-experts, and, second, the way a reasoning system will use this reliability to perform better results. The management of the reliability of knowledge units uses MKM (meta-knowledge model), a model we designed especially for this purpose. MKM is based on meta-knowledge such as belief, trust and reputation, about knowledge units and users. This meta-knowledge allows to compute the reliability of each knowledge unit. By using MKM, a reasoning system will be able (1) to filter the knowledge units for reasoning only with reliable knowledge, and (2) to better rank its results according to the knowledge reliability of the knowledge units involved in these results. In this paper, we present an evaluation of the benefits of this approach, in the context of a case-based reasoning (CBR) system which adapts cooking recipes. The results provided by the CBR system using the e-community knowledge without taking into account reliability are compared to the results provided by the same CBR system taking into account reliability. This comparison shows that users are better satisfied with results provided by the system which exploits the knowledge reliability. A qualitative analysis shows that the results returned by the CBR system which do not take into account the knowledge reliability and evaluated as bad by users are due to the use of knowledge units which are not reliable according to MKM. These results are no more returned if the knowledge reliability is taken into account. Moreover, using in priority the most reliable knowledge allows to rank a result at a better place, comparing to its rank in the system which does not use the knowledge reliability.