Prefered Method Of Network Redundancy (Server Connectivity)

There are several methods of offering network path redundancy. The basic decision for me has come down to:

Do it at Layer 2
Do it at Layer 3

At the moment we have layer 2 redundancy to each server. 2NICs on the server up with one having the IPs for apache/mysql/etc. each nic going into a seperate switch and the switches connected together, with 2 routers running VRRP to handle the gateway.
Everything is Vlaned.

So basicaly the switch redundancy is done by spanning tree and the IP redundancy is done with a process on the router/server to move the IP to the other router/server NIC if there is an issue.

I am thinking about going with 2 fully seperate networks. in differant subnets. where each router would have a gateway. the routers would talk to each other and they would speak OSPF or ISIS with the servers. This way I would move hosting onto loopback IPs on the servers and those loopback IPs would be advertised to the both routers through seperate networks.
Again everything would be Vlanned.

This has the advantage of getting rid of spanning tree which has caused issues from time to time. It would keep redundancy up and we could standardize on the routing daemon ran on the servers allowing all of the various OSs we run to have the same basic config for network redundancy. Where now each OS tends to have its own solution. This would make life easier from a config and troubleshooting point of view.