I don’t think it will ever be a practical reality. The main difficulty, of course, is “decoherence.” In essence any measurement tends to disturb a qubit’s superposition of states and normal particle interations in the universe act as measurements. To a degree there is also the fact that qubits, like classical bits can develop errors. Now with classical computing, dealing with errors is easy. You set up a couple million bits to represent a single bit. When you want the state of the true bit, you measure all the representatives, accept the majority, and force all the bits into agreement.
With qubits, things are a little trickier. If you make any direct measurement (before the end of the computation) you will get a value but lose the interconnection to all the other bits in the system. So they are looking at ways to detect relative errors without actually detecting whether the bit is in the ‘zero’ or ‘one’ state. But it is still a measurement. It still interferes with the superposition of states of the system as a whole. That is what I think will ultimately become an insurmountable problem. As you error-correct bits independently, they will become more and more independent, and less and less useful to quantum computation.