Like many of my posts, this is more of a sanity check question than a technical one, and, yes, I understand that some of this may be related to what makes good test subject matter versus what makes good networks. I've gone through a good portion of a few MPLS books and the SP videos, and have started labbing scenarios. The most interesting, so far, has been ways to provide customers with both MPLS VPN and internet access.
As unpopular as NAT is with the technical purists, if I was the customer and wanted an L3 VPN with a hub and spoke topology, it seems like a default route through a NATed connection (real interface or subint) in the global table, separate from their VRF routes, would be the best solution.
IRL, I have never heard a single request for L3 VPN at work, and frequently have requests for L2 solutions. This may have to do with self-selection via our specific customer base rather than what's actually popular in the industry at large, though, I realize.
To my way of thinking, it makes no sense at all (IRL, from the POV of a customer) to go through all of the trouble of setting up an L3 MPLS VPN between multiple sites and then announce your entire address space to the internet. This is essentially what route leaking a customer VRF to the global table is, correct? It also seems like, for the most part, companies are probably using RFC addresses internally so in many cases leaking their internal routes to the internet would be a non-starter to begin with.
What am I missing here?