We present the study of the gas mass fraction, $f_{\rm gas}$, properties of distant galaxy clusters observed within the XMM-$\Omega$project, based on a sample of distant ($z\sim 0.5$) clusters of median luminosity. These clusters were previously detected as serendipity ROSAT clusters surveys. We first study the $f_{\rm gas}$ profiles traced up to the virial radius in both Einstein de Sitter (EdS) and concordance cosmologies. We find that the mean gas fraction profile of our high-$z$ sample is similar to the local mean profile and in good agreement with numerical simulations predictions. This result confirms the self-similarity of the shape predicted by the simple scenario of structure formation up to high redshifts. We observe $f_{\rm gas}$ in the central part of distant clusters to lie below the local ones in the case of an EdS model, consistently with the observed evolution in the $\hbox{$L_{\rm X}$--$T$}$ relation, which is indeed found to be weaker than expected in a self-similar model. We investigate quantitatively this departure from scaling laws by comparing $f_{\rm gas}$ in our distant XMM clusters with those derived from the local sample at fixed scaled radii. We show that the internal structure is very complex, i.e. the variation of the gas mass fraction with radius, with temperature and with redshift reveals differences that cannot be described in a simple scaling scheme. The observed variations in the central parts are a clear indication that, in these regions, the argument based on the non-evolving $f_{\rm gas}$ with redshift cannot be used to infer the cosmological parameters, given the present (lack of) understanding of the gas physics in clusters. At the virial radius, where this argument could be valid, we find that the $\Lambda$CDM model is excluded at more than 3-$\sigma$.To search for other articles by the author(s) go to: http://adsabs.harvard.edu/abstract_service.html