The complementarity between coherence and entanglement or equivalently interference visibility in first-order or second order measurements is discussed for example in Birgit Dopfers PhD thesis ("www.univie.ac.at/qfp/publications/thesis/bddiss.pdf" ), unfortunately written in German around page 45 and 46.
It does NOT have anything to decoherence in the traditional sense. Just reconsider what single photon interference measures. It is a measure of spatial coherence of the light field which directly translates to a small spread in emission angle or equivalently momenta as seen at the position of the double slit. Or in other words the more point-like the light source is, the higher the visibility of the single photon interference pattern will be.
The reason for that is simple. Just imagine an extended light source and calculate the possible path differences from each point of the light source to the slits. These will of course translate into phase differences. The interference pattern behind the slit depends on the phase difference of the fields originating from the two slits. However, if there is already some phase difference introduces at the position of the slits, this will modify the interference pattern seen. So for an extended light source you will get a weighted superposition of all these slightly different interference patterns. If the spread in possible path differences is too large this corresponds to no interference pattern at all.
Under the conditions discussed in Dopfer's thesis, the minimal distance allowing to see a single photon interference pattern of perfect visibility is 770 mm.
The conditions for seeing interference in coincidence counting are rather different. The archetypical experiment is the one where the detector behind the double slit is placed behind a lens to get far field conditions and is not moved, while the detector in the other arm without any double slit is placed in the Fourier plane (each detector position corresponds to a certain k-value) and is moved around. Now each position of that detector corresponds to some specific momentum value and every photon detected on the other side will have a corresponding momentum value. There is typically no interference pattern behind the double slit (as discussed before) because the spread in momentum values is so large. However, as one now picks a certain momentum value by choosing a certain detector position in the Fourier plane, one also gets a relative count rate corresponding to the count rate one would see if one placed the detector behind the double slit at the very same position and fired a light field with the chosen momentum value at the double slit. As one moves the detector in the Fourier plane around, one picks a different momentum value and the count rate on the other side will change accordingly. If one moves further and further, the corresponding count rate will show minima and maxima according to the count rates one would see at exactly that detector position if one used light with the chosen well defined momentum. In summary one finds an interference pattern in coincidence counts.
Now why does the latter not work with spatially coherent light? This is almost trivial. As said before, spatial coherence corresponds to a small spread in momenta. As you now move the detector in the Fourier plane around, you scan exactly the whole range of momenta. One will find that the spread in momenta needed to see a single photon interference pattern is so small, that when you now scan the detector in the Fourier plane where you also scan the whole range of momenta, you will reach the end of the spread before you even reach a minimum of the coincidence count interference pattern. Under the conditions discussed in Dopfer's thesis the largest possible distance to see the interference pattern in coincidence counting is 106 mm.
The difference between 106 mm distance and 770 mm distance is huge and there is no region where you can get both. Note that this is not a consequence of the setup used. With other light sources or slits, you can change the numbers, but the upper distance bound for two-photon interference will always end up to be much smaller than the lower bound for single photon interference.
As an alternative simple handwaving argument, you can also think of the effective size of the light source becoming so small that diffraction from that point source destroys all correlations. Dopfer also gives this handwaving explanation, but in my opinion it is not a really good one as it works for sources having small size, but is not so trivial to translate into sources having small angular size (this is what you get by increasing the distance between source and slit).