Explanation/Reference:
I changed this answer to reflect CCDP explanation:

■ Layer 2 loop-free design: In this design, the access switches use Layer 2 switching. The links between the access and distribution layers are configured as Layer 2 trunks. The link between the distribution switches is configured as a Layer 3 routed link. An EtherChannel is typically used for this link to increase availability. In this design, there are no Layer 2 loops in the access-distribution block, which means that the Spanning Tree Protocol is not involved in network convergence and load balancing. All the ports are in the spanning-tree Forwarding state. Load balancing of the traffic from the access to the distribution layer is based on the First Hop Router Protocol (FHRP) that is used in this design. Reconvergence time in the case of failure is driven primarily by FHRP reconvergence. A limitation of this solution is that it is optimal for networks where each access layer VLAN can be constrained to a single access switch. Stretching VLANs across multiple access switches is not recommended in this design.
■ Layer 2 looped design: The Layer 2 looped design also uses Layer 2 switching on the access layer, and the links between the access and distribution switches are also configured as Layer 2 trunks.
However, unlike the Layer 2 loop-free design, the link between the distribution switches is configured here as a Layer 2 trunk. This configuration introduces a Layer 2 loop between the distribution switches and the access switches. To eliminate this loop from the topology, the Spanning Tree Protocol blocks one of the uplinks from the access switch to the distribution switches. This design is recommended for networks that require an extension of VLANs across multiple access switches. A drawback is that network convergence in the case of failure is now dependent on spanning-tree convergence that is combined with FHRP convergence. Another downside is limited load balancing. PVST root election tuning can be used to balance traffic on a VLAN-by-VLAN basis. However, within each VLAN, spanning tree always blocks one of the access switch uplinks.
■ Layer 3 routed design: The Layer 3 routed design uses Layer 3 routing on the access switches. All links between switches are configured as Layer 3 routed links. The advantage of this design is that it eliminates the Spanning Tree Protocol from the interswitch links. It is still enabled on edge ports to protect against user-induced loops, but it does not play a role in the network reconvergence in the access- distribution block. FHRPs are also eliminated from the design, because the default gateway for the end hosts now resides on the access switch instead of on the distribution switch. Network reconvergence behavior is determined solely by the routing protocol being used. Like the Layer 2 loop-free design, the Layer 3 routed design constrains VLANs to a single access switch. Also, this design does not allow VLANs to be extended across multiple access switches, and it requires more sophisticated hardware for the access switches.

This WAS answer !!!!!

Layer 2 between distribution and access layers, with a Layer 3 link between the distribution switches
-> Support Layer 2 VLANs spanning multiple access layer switches across the distribution switches Layer 2 between distribution and access layers, with a Layer 2 link between the distribution switches
-> FHRP for convergence, no VLANs span between access layer switches across the distribution switches VSS -> Convergence (FHRP) is not an issue The following are recommended best practices at the distribution layer:
Cisco Press CCDA 640-864 Official Certification Guide Fourth Edition, Chapter 3