Some years back, I worked for a large global enterprise that had MPLS and Internet, side-by-side. Internet was much, much less expensive and much easier (time wise and bandwidth wise) to obtain around the world.
Originally, the Internet links were purchased as a "warm" backup to "private" WAN links (originally p2p, then FR, then ATM and then MPLS) (for a while we also used some satellite links, but not for routine data traffic, just some "special" streaming video, over IP) and was considered it might be useful to direct bulk data transfers across it (while still in the role of backup).
My experience with the Internet links, which we in North and South America region used just like the "private" WAN links, was it usually performed as well if not better than "private" WAN links.
The "secret" to getting good usage out of the Interlink links was not to use the same Internet connection as both a "private" WAN link and for general Internet access. For the latter, we would obtain another Internet link for that. The reason for this, general Internet access creates an unpredicable load on the Internet link. This was crucially important, because we used QoS on both the "private" WAN links and on the Internet links to provide/manage SLAs.
I will mention, when we moved to MPLS, we also moved to a logical multi-point topology, which, again since we used QoS, really required QoS within the MPLS cloud, which the MPLS vendors provided (although only a subset of what our non-MPLS QoS supported [I predicted this may lead to breakdown of SLAs - and was later shown I was right]). To somewhat mitigate lack of MPLS QoS support, and other MPLS issues (like the two times a regional MPLS node became a black hole, not to mention occasional MPLS regional "brown-outs", we also used OER (later known as PfR).
MPLS vendor wise, we used AT&T globally, and Verizon (at least in North America). I don't recall there being another global MPLS vendor, at least back then. So, basically, at some larger sites, we allowed traffic to jump between MPLS vendors so that, generally, and globally, any site could reach any other site even if they lost connectivity to AT&T. (The mix of MPLS vendors also lead to some "special" handling of BGP, as the different MPLS vendors had different BGP topologies.)
