Same as SAS - matter of fact SAS wiring used similar connectors at one point.
Its like CX4 but 4 times faster! Parallel versus serial. Cabling is always cheaper if you can modulate with less wiring. SFP is 4 wires, i'd suspect QSFP is 16 wires. Fiber is one wire (each direction). 10gBASE-T is 8 wires. There is also now IPoTB (thunderbolt) coming soon to eliminate the costly nic when doing very short haul (think one mac server to 4 clients at a very short distance).
Or SMB3 which is multi-connection sometimes as well. without optics (lasers) it is hard to go faster for any distance which is why intel is working out using optics instead of wiring(traces) for communication on a motherboard and even between two chips.
It's likely if you wired your house with fiber, you could have gone from gigabit to 2 to 4 to 8 to 10 and on without rewiring at a short haul distance of between floors. Just change optics !
SFP cables are very touchy around 5 meter. I've had issues with passive DAC SFP+ cables only at 5 meters. Usually when using two ports. If its one SFP+ fiber and one passive DAC - no problem. Two fiber 300m no problemo. Seems to be reaching the limit with 10gbe DAC passive.
Remember passive SFP cables are just 4 wires with high quality copper with a ton of shielding (over each wire and over all 4 wires).
10gbase-T is a special bird since it uses a scrambler instead of a serializer so it has to encode the whole chunk (packet'ish) and send it. this is 8 times slower than a serialized fiber stream and probably 10 times more power to run it over 50meter of wire. And a ton of cross-talk
The real problem is generating data fast enough - some PC's don't have fast enough ram bandwidth to sustain 40gbe! let alone two ports of 40gbe yumminess!
how many hard drives would it take to sustain 40gbe at 100% random read/write pattern? alot!