How this seems to have been engineered...
One cable output provides instant access to 8 ports
1 Gigabit Ethernet port
1 FireWire 800 port
1 Thunderbolt port
1 3.5mm Headphone Output Jack
1 3.5mm Audio Input Jack
3 USB 3.0 ports*
Daisy-chaining ability of up to 5 additional Thunderbolt devices
Includes Cable-management channel.
* USB 3.0 ports data transfer at 2.5Gbps max
1 Gb Ethernet port at full speed
.8 Gb Firewire port at full speed
3 x 2.5 Gb 3 USB 3 ports, half speed each
9.3 Gb plus 700 Mb for audio use
Total is 10 Gb.
Even if the USB 3 ports could go full speed, two in use would saturate the link. Not stating this is an excuse for a $300 price tag. At least half of the cost of the device seems to be tied up in paying for the connectors.
That's not quite how it works though.
If you ignore the DisplayPort aspect, a Thunderbolt controller acts like a hypothetical Ethernet router with 4 downstream ports operating at 4 Gbit/s and one 10 Gbit/s upstream port. Devices connected to the downstream ports can operate at the full 4 Gbit/s, but may not be able to do so 100% of the time if the uplink is maxed out. In general, you can oversubscribe the hell out of that single uplink, and, with efficient switching, the end user will rarely ever notice.
To explain the analogy in more detail, the Thunderbolt controller in question here has connections for 4 PCIe 2.0 lanes on the back side and 4 Thunderbolt channels (2 per port) on the front side. Internally, the PCIe lanes are connected to a PCIe switch, hooked to a Thunderbolt protocol adapter, connected to a Thunderbolt switch. PCIe 2.0 nominally operates at 5 Gbit/s, however it uses 8b/10b encoding (10 bits are sent for every byte of actual data), which means 4 Gbit/s in reality, whereas Thunderbolt channels actually do provide 10 Gbit/s to the upper layers. Although Thunderbolt ports and cables are dual channel, one of them, at least in Apple's implementations, appears to be reserved for DisplayPort traffic.
I'd wager the Belkin dock uses 3 or 4 single-lane PCIe devices connected to the Thunderbolt controller to provide the bulk of its functionality: one 4-port USB 3.0 host controller (or two 2-port controllers), a single-port FireWire 800 host controller, and a single-port GbE controller. The audio is provided by a USB connected audio device which uses one of the 4 USB ports internally. The FireWire and GbE controllers are pretty straightforward, but the USB solution is worth discussing because of the note about it only operating at 2.5 Gbit/s.
This is a pretty blatant indication that the USB 3.0 controller is only able to connect to the Thunderbolt controller at PCIe Gen 1 speeds. Mind you, all of the discrete USB 3.0 host controllers that have made it into production thus far, regardless of whether they support 1, 2 or 4 ports, only have connections for a single PCIe lane on the back side. Since PCIe 2.0 and USB 3.0 are both 5 Gbit/s with 8b/10b encoding (i.e. 4 Gbit/s) interfaces, this works out quite well and individual ports are not limited. If you're aggressively using more than one port of a multi-port controller, well, YMMV. On some of the early silicon, however, that single PCIe lane was only Gen 1, or 2.5 Gbit/s.
Unless Belkin got one hell of a deal on some chips that have been lying around for a couple years, it would really surprise me if they didn't use a more modern 2 or 4-port solution. My money was on the xHCI 1.0 compliant Fresco Logic FL1100, which I believe is supported natively under Mac OS X 10.8.2 and later, and made an appearance on an ODM Thunderbolt docking solution from Pegatron at Computex last June.
We'll have to wait for a teardown, but either the USB 3.0 host controller is at fault here, or there's an issue with a PCIe switch not being able to negotiate a Gen 2 link on the port it's connected to. Both the GbE and FireWire controllers are likely Gen 1 devices, since that's more than enough bandwidth for them. I wonder if the Thunderbolt controller's integrated PCIe switch can't handle different rates on each of its ports.
