I have been looking at the SRND for Branch Lan design. In the section concerned for upto 50 users, The first suggested way is to connect 2 switches 1 on each port 0f a 2801 interface. I am confused as to how this works. If you have a voice vlan, a data vlan and a native vlan, then how will this work? If you put all the data on one switch and voice on another that will be fine, but I would have thought you would want to put half your phones on one switch and another half on the other in case a switch fails?
As to whether it makes more sense to span the VLANs, as is often the case, it depends.
With a router providing L3 between the two switches, you have all the "advantages" of L3 isolation between those two switches. This thinking is much alike what's also mentioned in the later design guides suggesting L3 within the access edge.
In this specific instance, since we're dealing with a small branch, and if we assume, all/most traffic is going to/from a WAN link off the 2801, i.e. no host-to-host LAN traffic, then perhaps L3 segregation of the two switches would be best.
If on the other hand, there was LAN host-to-host communication, and it needs to span the two switches, considering the performance of a 2801, I would think spanning at least the data VLAN would be better.
What's not always obvious is what the assumptions are about traffic flows in the design guides. One size doesn't always fit all.