I am not sure how you are determining your efficiency but let me try to clear up some misconceptions.
First off is the advantage of full versus half duplex. If you are running two machines and one session then their communication will be like a conversation. One talks, the other responds. The responses are typically based on what the other said, so you get minimum gain from full duplex. Each end has to make sure the other has buffer space for the packet being sent to prevent overruns and retransmissions.
Next you have the processing from application down the protocol stack to the NIC and then out the wire, and back up from the other side. The last bench mark I saw for essentially a data transfer (FTP) maxed out at 40% (40 Mbps) in a setup like yours. That was a couple of years ago, but I haven't noticed any major changes that would improve on this dramatically. In short, at 10 Mbps the transfer was waiting on the wire, but at 100 Mbps the wire is doing the waiting.
Other issues include packet size. A simple response in TCP is 50% overhead. A full size ethernet packet is about .1% overhead. Large packets are the preferred way to go. However, some applications were designed for older and less reliable circuits. These put an MTU cap at about 500 bytes to reduce the size of retransmissions. Other protocols have their own overheads to account for, and encapsulation adds to the overheads. MTU adjustments may be necessary at serveral points to get optimize the data transfer.
Even on 10 Mbps half duplex LANs, most application tasks only spend 10% of the total time in data transfer. The rest is in processing at either end. This analysis was done with scripted keystrokes to eliminate user performance variables.