Sorry very confused by your latest statements. If you want routing/communication between the communities why are you using PVLAN in the first place? Is the idea that communities can only talk to each other once they hit some Security point, like say a FW?
What is the subnet mask associated with the IRB and what is the subnet mask of the communities. Does a community know it needs to route (from an IP perspective) if it is trying to reach a different community?
Trying to figure out the big picture requirement, not just if IRB works with PVLAN, . . .