Rob,
All Ethernet running on twisted pair uses differential signalling. The tight twist of the pairs just about guarantees that any voltage induced on one line of the pair will also be induced on the other, making the difference between the lines within a pair stay pretty consistent. What is more likely is that the ring voltage could be fully induced onto a data pair, and some hubs might object to that (i.e. they may fry. I've seen it happen, although there was no way to know what induced the voltage on it). That being said, I would still not do it.
The problem is at the termination point. How do you break out pair 1 (for voice) from pair 2&3 (data)? There used to be some companies making adapters for this, but it was dubious whether they actually met cat 5 standards or not. And they weren't cheap. If you attempt to do this yourself, you are guaranteed to violate cat 5. Does that mean it won't work? Not necessarily, but you never know. Where I used to work, we had all kinds of bizzare things work fine, and some that were done to the letter that wouldn't (we had one run of cat 5 that had some kind of junk being induced on it, and it wouldn't work with our hubs). Even if it works, without using the pre-fab adapters, you end up with a mess at your outlet.
Stick with your original plan and run separate voice and data lines to each location. In fact, the last time I looked, the 568 standard called for two data and one voice to each location. Given how cheap the stuff has gotten over the years, why not do it? The 10 years that 568 was supposed to allow for are rapidly disappearing. And definitely do home runs, and terminate at a 110 block or cat 5 patch panel. This makes it incredibly easy to route just about anything to just about anywhere.
Kevin