The way twisted pair works, a DC offset really isn't a problem. A DC voltage will create EMF, and that will induce a voltage in nearby conductors. Because the DC is constant, the voltage induced on both conductors of a twisted pair will be virtually identical. If you add one volt DC to both conductors, the difference in voltage between the conductors doesn't change, therefore we have no ill effects from the DC. When POE shares conductors with the Ethernet signal, all you are doing is letting the signal ride on a DC offset. This is stripped off at the powered device, and does not change the signal.
An analog phone has a DC carrier, low frequency ring voltage, and low frequency audio. Of all the things that can share a cable with Ethernet, an analog line is going to be the safest.
Digital phones to the best of my knowledge are operating at well under 1Mhz. They would have a greater chance of causing interference, but still not likely to be a problem under normal circumstances.
Two data connections on one cable is the most likely to cause a problem, since you are putting two signals of the same frequency into such close proximity. It's like a mix between crosstalk and alien-crosstalk. 10Base-T would be less susceptible than 100Base-T. If both computers have a relatively light load you reduce the likelihood of problems.
AMP does have their ACO product which does allow you to split a cable for two 100Base-T connections. They were popular with government offices here in town back in the 90's. I wired some of those government offices so I suppose I'm as guilty as anyone else of splitting data cables. (I was young, didn't know any better, and I can blame the government.) I'm sure that product made sense before Gigabit, but I don't see why anyone would want to intentionally prevent their infrastructure from supporting Gig now.
At the very least, these ACO inserts were engineered for this specific purpose. If a cable must be split, this is how it should be done. Peeling back the sheath and splitting the pairs between two keystone jacks is absolutely going to have a negative impact. It might cause issues with additional noise, return loss or NEXT. You really won't know because you can't properly test the cable.
Cables are built with specific geometries for a reason, and when that geometry is compromised, that is when we have issues with EMI,RFI and crosstalk. If you kink a cable, the pairs are spread apart. This reduces the effect of the twisted pairs and that exact point is where noise will be induced on the pairs. If you have two Ethernet signals in one cable, and that cable is kinked or crushed, the crosstalk between those two signals will be much worse than if they were in separate cables. Physical separation is important.
There are situations where splitting one cable for voice and data is a reasonable solution, even if it isn't the last resort. However, I've never seen a situation where a switch couldn't be used rather than splitting a cable for two Ethernet devices. Either way, this is a decision that should be left up to the trained professional. I'm not trying to tell the pros anything they don't already know. This is aimed at anyone who may simply gloss over the cautions, and decide this is an excellent way to save money on cabling. Being cheap is not an excuse to split a data drop.
