Be very careful with that idea, unless you know EXACTLY what you are doing.
A) Adresses are in VRFs
It will cause no harm if these sets of loopback interfaces are on PE routers (or multi-vrf CEs) AND are part of different VRFs each (since were' talking about MPLS-VPN, here). In other words: IF each customer (VRF) gets their own set of
loopback20|30|40, AND IF each set is mapped to its respective customer's VRF, then overlapping addressing is perfectly acceptable.
The overlapping addresses will be picked up by the VRF on the PE, redistributed (by virtue of
route target export/import) as unique vpn4 prefixes (by virtue of individual route distinguishers) across the MPLS-VPN domain. That just goes to demonstrate what VRFs and MPLS-VPNs are all about.
If you deploy multiple identical sets of loopbacks/addresses within a single customer's VRF, then you get C) (see below), but within the customer's VRF.
B) Addresses in GRT, interfaces isolated
IF these loopback interfaces become part of the Global Routing Table, AND IF these loopback interfaces never become part of any (dynamic) routing protocol, this will cause no harm. Typically, becoming part of a routing protocol would be by virtue of a
network statement in OSPF or EIGRP, or by
redistribute connected in the given routing protocol. In that case, each router having these interfaces will just happen to have them (appearing locally as "connected route") and tell no one else about it. These isolated addresses/interfaces would be reachable only locally from/within the router which has them.
C) Addresses in GRT, enabled for dynamic routing
IF these loopback interfaces become part of the Global Routing Table AND will be part of the (GRT's) routing protocol, these addresses will become anycast addresses. Packets towards one of these will be forwarded towards the "nearest" ("near" in terms of routing cost of the given routing protocol) instance of the given loopback interface resp. address. That may or may not be intentional and have some unexpected consequences for traffic to/from these addresses. Anycasting with loopback interfaces is something rather special, but a perfectly valid thing to do (one application is "Anycast-RP (with MSDP)" for multicast routing).
I want to configure a set of loopback interfaces on every PE router
for routing traffic between different VRFs.
This raises the question: Why configure separate VRFs in the first place, when you're going to connect them back together again?
If you need inter-VRF connectivity (for example to make central services available to the customers), then don't to that on the PEs. Rather set up a dedicated "Service PE" pair and connect a firewall cluster (with 1 (sub)interface per customer/per VRF) to it, and let it handle the selective connectivity needs between VRFs (and possibly NAT, to handle any possible overlapping address situation).
There's always the possibility of route leaking from VRF to VRF, but that is a can of worms I suggest you open only when really needed.