I would say you must ALWAYS calculate delta-H-rxn from the PES. The reactants and products must be equilibrium structures with 0 imaginary frequencies. The thermally corrected enthalpy difference between those (products - reactants) is what we call the delta-H-rxn. The IRC typically does not go far enough towards the reactants or products to reach their equilibrium structures.
I think it should be done from IRC curve because in an IRC curve there is guaranteed only one transition state, which guarantees that the curve is for an ELEMENTARY REACTION and not for ANY OTHER reaction or SERIES OF RECATION. This means that the delta-H-rxn we would get from an IRC is guaranteed for an ELEMENTARY reaction step, while delta-H-rxn we would get from PES is not necessarily for an ELEMENTARY reaction as PES can have peaks and troughs which means many possible transition states and intermediates .
I respectfully disagree. You must know what the reactants and products are (regardless of how many steps may be involved in between). The enthalpy of reaction is based on THAT enthalpy difference. For example, the Fischer-Tropsch or Water Gas Shift reactions involve many catalytic steps. However, the enthalpy of reaction refers to the starting materials and the final product of interest. We don't need to know anything about the intermediate steps.