Not sure why you need to segregate anything when BPGA has already done it for you. The groupings were done based on distribution in multiple genomes, which is a broader relationship criterion than simple sequence similarity.
For genes that have paralogs, one of them may be in the core group, and others are in the accessory groups. Those paralogs could still retain relatively high percent identity and coverage, and most definitely E-value lower than 1e-4. If not universally present in multiple genomes, they will go into accessory group.