The whole Panasonic code is the repeating code. So the best you could do with that second learn is get the same as the first, which you already have.

Apparently the delays between macro steps are too long.

There are protocols where the repeat part differs from the initial part and the method you describe will sometimes do as well as you could do by editing the hex into two parts (if you understood the simplest parts of Barry Gordon's document on Pronto Hex format). But for this protocol there aren't two parts to split it into, so neither strange learning methods nor editing the hex can create any repeaqt part that works better than the original.