I am not an expert in compound-specific isotope measurements, but I do know there are several other factors to consider besides H3+ and nonlinearity in the detectors. As mentioned before, incomplete conversion of your alkanes to H2 would cause an apparent fractionation effect correlating to peak height, if a greater fraction is lost at lower sample mass. If you can discount the reactor, the GC column may be retaining sample that does not get integrated into your peaks. This would likely cause poor chromatography, which can effectively bias your samples based on peak size. Even if your peaks are sharp, some sample may be retained on the inlet, which also would cause a similar-looking delta-to-size correlation. I'm not saying you should give up on the goal of wide size dynamic ranges, but it may be easier to just run standards that can correct your data given the potential biases, or pre-screen samples for manual dilutions or concentrations.