Sorry I'm late to this thread. There are good reasons to separate the capsules at least a little, especially if they're cardioid.
If you're going to use coincident placement for whatever reason, though, definitely aim the capsules apart from the center line, not toward each other. When two microphones are placed close together and aimed toward a common center line, they partially block one another, which distorts their pickup patterns. Unfortunately there are still some articles and textbooks that show microphones placed "head to head" for coincident pickup. It's much better to place one capsule (or one microphone) directly above the other and aim them both outward from their common center. Ideally you want a common center line for the two diaphragms--an arrangement which many "coincident stereo microphones" create automatically (but see below about the dual-diaphragm capsules which are usually used in such microphones).
Back to why separate the microphones at least a little: If you don't absolutely need mono compatibility, recordings will generally sound and feel more spacious in playback over loudspeakers when the stereo effect is created partly from arrival-time differences between the channels in addition to amplitude differences. The explanation for this is fairly technical and has been discussed in many places including this board.
I would make an exception to that statement for the Blumlein method with coincident, crossed figure-8s at 90 degrees, though. In those (unfortunately relatively rare) recording situations where the geometry of that type of setup is appropriate to the recording environment, Blumlein tends to work very nicely. The pickup regions of the two microphones are so strongly independent of each other that the signals that the two mikes pick up will automatically have low correlation between the channels--which is what you want.
That is definitely not the case with cardioids, however. A cardioid pattern is relatively broad. If you set up a coincident pair of cardioids and aim them 180 degrees apart (back to back)--thus getting the greatest possible difference in what the two channels will pick up--about half of whatever you pick up in either channel will be in the other channel as well. As a result, when you play back the recording, even for sounds that originated far to the left or right in the real world, the apparent source of the sound never quite reaches either speaker position. Instead it tends to "puddle" in the middle; there is always a strong central (i.e. mono) component to the recording. For this reason I feel that X/Y with cardioids at 90 degrees is a poor choice of a "cookbook recipe"; I would suggest starting at 120 degrees instead (of course, always adjusting this for the actual circumstances).
This effect is even worse when the cardioid microphones use dual-diaphragm capsules, as in most switchable-pattern microphones, because such capsules have a true cardioid pattern only in the midrange. At low frequencies the pattern tends to "bloom outwards"--it becomes a wide cardioid. That's nice for studio vocals, since it tends to warm up the sound, but for coincident stereo recording, at low frequencies the pickup will be even closer to pure mono than what I described above. This is a serious problem with coincident stereo microphones even from such high-end manufacturers as AKG and Neumann, who use dual-diaphragm capsules to provide multi-pattern capability. X/Y cardioid recordings should be made only with single-diaphragm cardioids (the most common type of single-pattern cardioid).
Anyway, putting some space between the mikes helps to reduce all these problems.
--best regards