(I knew you were being facetious; and still like 2 and a half.)
As to mixing L2 and L3, OP makes that a requirement. Otherwise, there would be no need for the one global L2 VLAN (and something we try to avoid with L3.)
The real question is whether to use two links or one. Either can work. Personally, I would rather use a 2nd link for additionally bandwidth and redundancy in a Etherchannel configuration. Assuming the 2nd link is even currently available. (If STP is involved with dual parents, you would need either 2 uplinks, my method, or 4 uplinks, dedicated method.)
However, for those who find it easier to understand and maintain dedicated links, that's certainly a valid consideration.
(For example I worked on one project were we needed to span OSPF areas between a pair of 7500s. I configured multiple subinterfaces on a FastEthernet port. Another engineer found that complex, so he found some multi-port Ethernet cards, and dedictated one port to each OSPF area. 7500 had spare card slots and user traffic didn't normally transit the links, so although I was happy with the subinterfaces, not a problem to do it his way either.)
As to impact of loop issues, if you configured it wrong in either case, you'll likely know it when it happens. Often a loop is more of a problem than just link saturation, and that might be addressed by port QoS. Personally, I would worry more about something like STP configured correctly to break a loop rather than whether a dedicated loop can handle the looped traffic. Again, though, if those maintaining the network want to trade off a link for, what they consider, less complexity, and they have considered other tradoffs, feel free to do so. There's often many ways to deliever the same results.
