You can use both but it depends on your design.
So a very common design (before stacking, VSS etc) was to have your distribution switches interconnected via a trunk link and then each access switch would have an uplink to both distribution switches. This meant that per vlan there was a loop so one of the uplinks from each access switch would block and this is why GLBP could be seen as inefficient because with GLBP either distribution switch could be the gateway for the client but if the uplink to the gateway was the blocked link then traffic has to take the long way round ie. to the other distribution switch then across the interconnect.
But you can design your network to work around this although it does come with a limitation. The idea is you interconnect your distribution switches with a L3 link instead. The access switches are still using L2 uplinks. Because the interconnect is L3 you now do not have a loop so both uplinks from the access switch can be forwarding which means GLBP now works efficiently ie. whichever distribution is the gateway for a client the traffic can be sent direct.
The limitation, and it is a big one, is that a vlan cannot be on more than one access switch. This limitation does not apply to the first design I outlined. The reason for this limitation is that if the same vlan is on multiple access switches then you create a loop via the access switches even with a L3 interconnect via the distribution switches. If you draw it out you can see it more clearly. So then you are back to the access switches having to block uplinks etc.
Hope that makes sense.
Jon
