Score with equalisation. Not a trivial thing to compute, but in the modern world, one where just about every AVR has the ability to provide equalisation, not unreasonable. This would emphasise the balance across the polar response and de-emphasise flatness of response. But smoothness would still need to be included where it is not reasonable to tweak out irregularities. Something to ponder.
1) Not everyone uses the same EQ, REW has 10 bands, Audyssey XT32 has thousands, DIRAC has near infinite.
2) You ideally EQ the predicted in-room response and not the on-axis or listening window. However, we don’t really know how to compute how altering the PIR will effect those curves.
Looking at the DI curves and the normalized horizontal directivity is pretty much what you will have to look at to determine how easy it is to EQ a speaker.