I've recently discovered that twinaxial cables can have very high transfer speeds of 100gbps. Even the ancient technology of RG-6 can get gigabit speeds with moca. The thing I don't understand is why aren't purpose built cables such as CAT6 with the additional conductors faster, and why do we even bother if we could be getting 100gbps with a two conductor cable? Any advantage besides POE?
There are two factors:
The propagation speed (or velocity factor VF as a fraction of the speed of light c0) depends heavily on the permeability of the cable's dielectricum. Essentially, the copper only guides the electromagnetic field that propagates through space. The VF is calculated by VF = 1/sqr(k), with k being the permeability of the cable (conductor and insulator/dielectricum combined). Coax or twinax cable can have very 'fast' insulators, with up to and even more than .9 VF.
Cat-5/6 twisted pair has a VF of only .65 (Cat-7 is around .75), so the electrical signal simply takes longer to cross the cable. Given the short lengths of especially twinax cable this is pretty negligible though - for 15 m, the maximum twinax reach, the difference is ~20 ns.
A much more significant effect comes from the line code. While twinax works with simple 64b/66b (for most port types) which has next to no encoding overhead, 10GBASE-T requires elaborate encoding which causes an port-pair overhead of .8 to 2.5 μs (depending on the generation and grade of the port hardware).
That encoding overhead is also the reason why fiber is considered faster than twisted pair even though the signal itself might actually propagate a tiny bit slower.
Speaking of 100 Gbit/s - there's only the short-reach 40GBASE-T for twisted pair and that's the end of the line there (permanently). Everything faster uses fiber or (very short) twinax/DAC/-CR.
 10GBASE-T uses Reed-Solomon forward error correction and Tomlinson-Harashima precoding, producing considerable overhead (for Ethernet), in addition to PAM-16 modulation with DSQ128 "checkerboard-pattern" symbol selection.
Simply put, twinaxial cables allow only very short distances, under 10 meters.
And they are used in networking, for example both 40Gb and 100Gb standards support twinaxial cables... ...up to 7 meters.
Due to this limitation they are mostly use in Direct Attach cables.
(outside of networking, they are used in USB3 or Display Port cables for example).
The background is the signalling bandwidth of each technology. Here is the baud rates at the introduction of each technology:
Twinax 30MHz. Although long-haul phone systems achieved about 45MHz.
Multi-mode fiber 160MHz.
Single-mode fiber 1400MHz.
Now data rates are much higher these days, but these figures give you a good idea of how hard a technology needs to be worked to achieve a desired speed.
With UTP that work means increasing the diameter of the wire (Cat5 100MHz), enforcing the twist and separation of the wires (Cat6 250MHz), and installation practices particularly bend radius (Cat5,6), impingement on the cable (Cat6) and termination (Cat6A 500MHz).
There has been a similar journey for twinax, with conductor distances becoming more controlled, the dielectric material between the conductors being improved, and bend radius increasing (usually done by decreasing the diameter of the coax). A side effect of this has been to make twinax physically fragile.
Multi-mode fibre has had a series of improvements, from FDDI (160Mhz), OM-1 (200MHz), OM-2 (500MHz), OM-3 (1500Mhz), OM-4 (3500Mhz), OM-5.
Single-mode fiber hasn't seen much improvement, as the single-channel use of that fiber is constrained by the bandwidth of the laser transmitter and receiver system rather than by the bandwidth of the cable. The major changes have been to allow more channels, by better control of the doping of the fiber, which removed an band of high attenuation called the "water peak". Fiber channels are conventionally 100000MHz wide, with conventionally 80 channels per fiber pair. For the most recent technolgoies, adjacent 100GHz channels are formed into 400GHz channels.
In general, the higher the baud rate, the lower the distance. As the cable gets longer the signal looks more and more like random noise. Higher-speed changes to the signal obviously start out looking closer to noise than slower signals. So one of the challenges of improving bitrates has being preventing the length from becoming too short to be useful. There's a few magic values from common practice: 10Km is a useful length of metro fibre; 2Km is a useful length of campus fibre; 100m is the maximum distance of UTP installations in buildings; 30m is a useful length for interconnecting racks; 5m is a useful length for interconnecting things within racks.
On top of these raw signalling rates (technically: baud rate) there is a modulation applied to transfer the digital bits to a analogue signal. These modulations have both electronics tasks (preventing DC bias in the signal) and data communications tasks (making the received signal distinguishable from the noise floor, coding the binary into physical, framing). These modulations can be simple or complex: complex modulations require more work by the receiving electronics, making them slower and using more power; but simple modulations may not achieve the desired bandwidth.
We can also 'cheat' and use multiple transmissions:
In a UTP cable there are four pairs of parallel same-length transmission lines. 100Base-T uses two pairs -- one pair for receiving, one pair for transmitting. 1000Base-T uses these four pairs as four individual transmission lines. 10GBase-T uses a complex modulation to drive all four transmission lines in an inter-related way.
For multi-mode fibre it's easy to build four parallel same-length pairs. The "MPO" connector allows these pairs to be treated as if they were a single cable.
For single-mode fibre four lasers can drive four different frequencies of light ("channels") across one pair of fiber.
Now for 10Gbps twinax we can put out a simple modulation across about 7m of cable. So that's a useful distance for working within a rack. But not a useful distance for longer runs. We could amplify the signal for greater reach ("active" cable). But then the reasons ethernet moved away from coax to UTP start to tell. At the top of that list: coax patching is expensive and unreliable.
Even within a rack twinax can be so annoying that as the price of multi-mode fiber falls, fiber is the more attractive choice. Twinax is fragile, the bend radius is too large, kinks are permanent. You can get non-fragile twinax, but its diameter of 8mm makes it unsuitable for high-density cabling, but high-density cabling is very much what we want to do for lengths <7m.