Note that BioPAX files were generated using Pathway Tools (PT) and PT has a known bug:

There are some pathways for which PT has trouble mapping the metabolic network representation to BioPAX level 2, and some for which PT has trouble mapping to BioPAX level 3, and those sets are not necessarily the same (because the models are quite different).  While theoretically, there is no reason why these pathways shouldn't be able to be represented in level 3, it looks like PT code has trouble with pathways that involve ferredoxin and generic electron acceptors, just due to the way they are represented in PT.

As an example, one of our users made the following analysis with MaizeCyc 2.2:
* The BioPax level 2 owl file contains 469 pathways.
* The BioPax level 3 owl file contains 429 pathways.
* The number of pathways that are in both files is 418.