Do you have per-function loopback interfaces on your routers? Why (not)?

In some customer environments I've come across routers with lots of loopback interfaces:

• One for the IGP's RID
• One for in-band management
• One for MSDP peering
• One for sourcing NTP client traffic
• One for iBGP peering
• etc...

This CL slide shows a modest example of the phenomenon.

What's the advantage of doing things this way?

I generally create an additional loopback interface when it's really required:

• a new VRF that doesn't have a loopback, but could benefit from it
• a service that's known by IP and which I might want to move later (like an NTP server or GRE endpoint.)
• an instance of an anycast service (these flat-out can't overlap with a unique-per router address.)

I'm guessing there's a philosophy here that I'm missing out on.

Enlighten me?