@MHM Cisco World's example of using IRB for a FW is good. I too thought to mention it, but not so sure you would actually see this too often in practice.
What also has to be kept in mind, the "price" of router ports vs. switch ports (which interconnect L2 segments all the time) and/or possibly other techniques to not only use something like IRB for interface redundancy, but something like Etherchannel which can increase bandwidth between devices too.
Consider, you could use IRB to have two L2 links from a switch to the router. This provides interface and link redundancy between the router and switch, but it also creates a L2 loop, which can be broken by STP. If we're so worried about redundancy, what about the router and/or switch itself failing? Again, Etherchannel provide the same redundancy, but doesn't rely on STP and, in theory, potentially doubles bandwidth between router and switch.
BTW, in that reference, Appendix B example joins two VLANs, which usually negates the reason for having two VLANs. but, as the FW is the router, traffic between the two VLANs is subject to different FW rules. A good example, but I suspect for such an actual usage case, you're more likely to see a physical FW. (Still, good example!)
