Oh yes, I definitely messed up my calculations. Half life is (inversely) proportional to the reaction rate itself I guess (I were associating it with the exponent, because half life is usually in the exponent), so the half life itself would be proportional to exponential reciprocal of temperature (e^(a/T)).

This means significant differences in HL for minor temperature variations under e.g. Arrhenius model (consistent with your graph I think). To extrapolate it some parameters need to be estimated though, which sounds interesting, I'll get around to that later...