The timescape cosmology has been proposed as a viable alternative to homogeneous cosmologies with dark energy. It realises cosmic acceleration as an apparent effect that arises in calibrating average cosmological parameters in the presence of spatial curvature and gravitational energy gradients that grow large with the growth of inho- mogeneities at late epochs. Recently Kwan, Francis and Lewis have claimed that the timescape model provides a relatively poor fit to the Union and Constitution super- novae compilations, as compared to the standard CDM model. We show this conclu- sion is a result of systematic issues in supernova light curve fitting, and of failing to exclude data below the scale of statistical homogeneity, z < 0.033. Using all currently available supernova datasets (Gold07, Union, Constitution, MLCS17, MLCS31, SDSS- II, CSP, Union2), and making cuts at the statistical homogeneity scale, we show that data reduced by the SALT/SALT-II fitters provides Bayesian evidence that favours the spatially flat CDM model over the timescape model, whereas data reduced with MLCS2k2 fitters gives Bayesian evidence which favours the timescape model over the CDM model. We discuss the questions of extinction and reddening by dust, and of intrinsic colour variations in supernovae which do not correlate with the decay time, and the likely impact these systematics would have in a scenario consistent with the timescape model.