They HAVE figured it out:
http://www.apogeedigital.com/products/symphony_performance.php
What JFreak has been saying is absolutely true; TDM systems have no additional latency added when plugins aren't in the signal path. When TDM plugs are added, there is a minimal amount of latency added. RTAS plugs are a different story.
But Apogee has created a native system that is so well integrated with the host that it can run at 1.6ms, analog to analog, through plugins. That's TDM territory. Other native systems can't touch this, especially not Firewire-based systems. FW systems HAVE TO have additional safety buffers to avoid problems at low latencies. That having been said, with a modern Intel Mac running a FW interface with well-written drivers (MOTU comes to mind), you can work pretty comfortably at 64 samples (64 in + 64 out = 128 total = approx 3ms)...I do this right now. Of course, I get added latency when I add plugins, but I get around this by direct monitoring and/or leaving plug-ins OFF until mix time. This leaves me with less flexibility than a TDM or Symphony-based system, but it gets the job done.
The Symphony system is a great proof-of-concept; native systems CAN run with latencies like TDM. But that will require a close collaboration between DAW, interface, and computer makers, and will require these people to value latency over 'frills'.