For the power side, there are multiple tradeoffs at stake, and most of them come out of Ohm's law: (V) = (I) * (R), where
(V) = potential, in volts
(I) = current, in amps
(R) = resistance, in ohms
...and its sibling power equation, (P) = (V) * (I), or by algebraic substitution, (P) = (V^2) / (R) = (I^2) * (R), where
(P) = power, in watts.
Presently, the power consumed within a PC is driven by potentials of 12V and under. To get more power, either (V) or (I) must be increased. If we increase (V), we quickly run into safety and durability issues because DC makes for some really prolonged, fat arcs when it is interrupted or finds a short circuit path, and higher voltages aggravate the problem. So, we have generally kept (V) low, and increased (I) when necessary.
However, since the conductor carrying (I) has its own value of (R), and (P) losses in the conductor are proportionate to (I^2), we have to make the cable length (L) shorter and/or make the conductor thicker, because losses in the cable cause heating and a (V) drop across its length. Likewise, the magnetic field (B) generated around the conductor is proportionate to the magnitude and rate of change in (I), and any interference effects transmitted to other conductors are proportionate to the strength and rate of change in (B) and the parallel (L) of the conductors. Although DC should theoretically have a continuous current flow, switching devices actually deliver and draw power in dozens of high-frequency pulses. These pulses are buffered on both ends, but a substantial ripple current cannot be avoided, so (B) is changing to some degree, and this can be transmitted to data-path conductors, modifying their signal. The catch is that (B) decreases exponentially with distance, so two conductors in the same cable jacket may have significant interference problems while two conductors lying a couple centimeters from each other might be perfectly fine.
A different-but-related set of problems is being encountered on the data conductors at the same time, as other posters have already detailed. And there's also the inevitable question of what happens if a power and a data pin short to each other -- can both devices survive the initial fault and shut down quickly enough to avoid damage?
The tl;dr is that as power capability increases on the power conductors and data bandwidth increases on the signal conductors, there are multiple tradeoffs being played against each other. Sometimes it makes sense to combine both functions in one cable and sometimes it does not.
He who laughs last, laughs first next time.