Here's my own take on things... we should not be talking about NACELLES at all here... the real issue is "coil sets."
Also... note that this is specific to LINEAR warp nacelles (what's been seen, canonically, since TMP, but which is arguably NOT the configuration of the pre-TMP warp systems).
For instance, consider the 1701-D. It has two nacelles, but it has FOUR coil sets... two nacelles, each of which has a top and a bottom row of coils.
Now, a single coil set creates forward movement by enveloping the ship in a subspace field, and by creating additional "nested" subspace field bubbles. Each coil creates a bubble, and the then next aftmost coil creates a bubble... and since the coils are arranged linearly, the compression of the subspace fields aftwards is much greater than that of the same fields forwards of the nacelle's axis. This creates a "warped" subspace field bubble... the bubble wants to go to it's lowest-energy state (where it would be perfectly symmetrical) but it's being forcibly created as a non-symmetrical bubble.. and that attempt to "equalize" is what gives the COIL SET some forward momemtum.
The thing to consider here is that each coil set can then, by definition (if you accept my explanation, obviously!) only provide a simple linear "straight ahead" (or, arguably, "straight behind" if you reverse the sequence of coil activation) "thrust vector." (Yeah, it's not really thrust, per-se, but you get the idea.)
So, fine... with one coil set you can go forward or back. But you CAN'T TURN!
So, let's say that you then have a system to put the sets of coils slightly out of sync with each other. Say you have a single nacelle with two sets of coils (sticking with the 1701-D example). You could slow the top coil set relative to the bottom coil set, very slightly, and as a result you could steer upwards. Or increase the sequencing speed of the top coil set relative to the bottom coil set, and steer downwards. But you can't turn left or right, and you can't roll.
Carrying this forward... with TWO nacelles (and thus four coil sets), you get the ability to steer up, down, left, or right, simply by adjusting the timing of the various coil sets.
You can also roll by adjusting the left nacelle to pitch up or down, and the right nacelle to pitch in the opposite direction.
And it makes sense that the further apart the coil sets are, the more effective your ability to steer in that axis would be. Thus, sticking with the 1701-D example, the ship should be able to turn left and right while at warp very efficiently (since the nacelles are spaced far apart) but would be a bit more sluggish in pitching up or down. It will also be able to roll, roll reasonably effectively... more so than it can pitch, but less than it can yaw.
Make sense?
So why keep two coil sets in a single nacelle? Well, in the case of the Stargazer, we see that some ships don't do that. But it seems to me that you're going to get the highest EFFICIENCY OF OPERATION when pairs of coils are closely coupled (meaning that they're adjacent, that the fields are able to merge very easily without much wasted energy, etc). In other words, the Stargazer might be very MANEUVERABLE at warp, but might not be particularly EFFICIENT at warp.
The two-nacelle approach works nicely, because it's balanced... you have one direction where you're very manueverable, but you are still pretty efficient.
And the one-nacelled approach would be much different... basically, you'd have four sets of coils inside of one housing. This is the approach you'd take for a ship that needed to be very efficient but didn't really need much in the way of FTL maneuverability. It could go like a bat outa hell in a straight line... just don't ask it to turn on a dime!
Anyway, that's been my theory for years now... it's consistent with everything we've ever seen on-screen, has no disproof (I don't take magazine articles to be "canon"), and makes good scientific sense (within the fictional "science" we have to work with here, I mean).
Now, there may be other ways to distort a subspace field and allow steering besides using sets of coils as I describe above. For instance, the original-series nacelles may not have been "linear" at all... I've always assumed that the white sphere at the aft end was a bubble-generator,and that the inside of the nacelle had at least one similar device internal to the structure, in-line... with those creating the forward/aft momentum, and probably with a couple of offset smaller bubbles in there as well to provide lateral adjustments.
Now, that means that, within an "old style" nacelle, you might have two (originally) or three (tos-series-version) "fore/aft" bubbles, and a pair of smaller "up/down" bubbles in each nacelle, and you'd steer left/right by adjusting the nacelles (similar to above). In a one-nacelle version, you'd just have four little bubbles (up/down and left/right) inside a single nacelle housing. The "little" bubble generators would be so much smaller than they would barely impact the construction of a nacelle at all.
So... that's my view of how these things work. Attack!
