A comment about coherence in general
I find the definition of coherence as some sort of "unrelated phase" problematic for a couple of reasons:
- This formulation somewhat implies that coherence is discrete, i.e. there is incoherent and coherent. Of course that is not true, you can have a continuum partially coherent states. But what quantity are you going to write down to quantify the degree of coherence? The "unrelated phase"-quantity?
- It does not emphasize that coherence is really about statistical correlations of observables.
Coherence is really defined by the following quantity:
$$g(r_1,t_1; r_2,t_2) = \frac{\langle E^*(r_1,t_1)E(r_2,t_2)\rangle}{\left[ \langle|E(r_1,t_1)|^2\rangle \langle|E(r_2,t_2)|^2\rangle \right]^{1/2}}$$
(see also wikipedia). So it quantifies the correlations of observables at different points in time and space of some statistical ensemble/time-dependent field.
IMO this removes all conceptual difficulties of coherence, but might be difficult to deal with in practice. The model that is described in the question is a perfect example: to compute the degree of coherence we would have to find the average in the expression above. For classical electromagnetic waves this involves an average over the ensemble that produces the field (1), for QFT or even just quantum mechanics it would also involve the expectation value of the the quantity measured (e.g. electromagnetic field).
Relation to the Question
In this picture there is no conceptual problem for few photon systems. Of course it might be practically hard to actually evaluate the averages, in particular write down the expectation values for the multiple photon states.
(1) For non-ergodic ensembles one runs into difficulties here and has to use a time average instead, which might be even more difficult to do in practice. I talk about the ensemble here because I find it more intuitive, but there is no difference with regards to the question.