Hello Kostas
This is an excellent question and it reveals some of the nuances of EIGRP.
When using Ethernet as the layer 2 protocol, holdown and hello timers for EIGRP don’t have to be consistent over the whole AS. Each EIGRP neighbour relationship must have its timers configured correctly, but those can be different from relationship to relationship. However there is one case where timers must be configured the same, and this is the case you describe above.
This case is when you have multiple routers on a network that form adjacencies. All those routers must have matching hello/dead timers or the adjacencies will flap. This is a common problem with EIGRP in a frame-relay hub and spoke topology like the one you are describing where an NBMA PVC does not support broadcast. In this case, like you said, the point-to-point subinterface will use the hello/dead timers of 5/15 whereas the non-broadcast PVC will use 60/180. This will cause the hub to have adjacencies with neighbours with different timers on the same physical network thus causing flapping adjacencies.
Therefore you must tweak the hello/holdown timer to be the same on all links.
As for the loopback interfaces, no you don’t have to change the timers on those interfaces because as I mentioned before, you can have multiple interfaces with different timers in the same AS. Secondly, timers on the loopbacks will not affect the whole topology anyway because there are no other EIGRP routers connected to those to require timers.
I hope this has been helpful!
Laz