I was under the impression that QoS is really only applied when the link is saturated
Well that's true if you also realize a link is saturated as soon as there's even one packet waiting to be transmitted.
My question being what is the advantage of making one priority and the other bandwidth?
Priority class queued packets are dequeued before any other queues. I.e.they are always sent next.
The shape average applied to all traffic?
Cannot tell from what you've pasted unless policy-map QOS-MPLS-OUT was meant to be QOS-MPLS-OUTBOUND. If so, yes it applies to all traffic.
If the gig interface is physically running at 100 Mbps, there's no need for the shaper.
If the gig interface is physically running at gig, a shaper is probably needed, but many Cisco shapers only appear to "count" L3 while provider CIR are usually counting L2. If this is the case, you're shaper's rate is likely set too high to be very effective.
PS:
BTW, in general, I recommend against using random-detect, unless you're a QoS expert. Additionally I would even more recommend against using it with fair-queue as you're doing in your class-default QOS-MPLS-100M policy class.
