Before you invoke "perfect detector" to support your belief in _future_
violations, you need to establish what would "perfect detector" detect?
Free EM field quanta (i.e. Dirac photons)? On what principles would
such detection work and does it violate QED cross-sections for
electron-photon interactions? Do you know of any design, even merely at
the conceptual level (but still respecting the QED and the known
physical constants)? And that is just the beginning of your problem.
Your "prefect detector" would not only need to count D-photons (which
depend on Fock base & on reference frame) with sufficient efficiency,
but it would need to know what the other three detectors of the Bell
EPR setup have shown in order to decide whether it has detected the
D-photon. Check, for example Ou & Mandel 1988 experiment [1], which
includes derivation of the QO "prediction" of Bell inequality
violations. In eq. (4) of [1], they restrict the correlations to a
particular 2nd order Glauber function G2(x1,x2) -- that means your
"perfect detector" (which would show violations before any data
filtering or "fair sampling" extrapolations) would need to know what
the other three detectors have decided, so that, for example, it can
report non-trigger when the other detector on the same side has
reported a trigger.

Namely, the G2(x1,x2) of their eq. (4) doesn't represent all the
scattering events (and their amplifications observed as photocurrents)
that are possible with their PDC source i.e. the eq. (4) is _not a
prediction of what will happen_ (even just statistically) in the
experiment, but merely that some post-selected subset of the events
(based on the knowledge of all 4 detector results!) will have
properties, such as the particular dependency on the experimental
parameters (e.g. the angle theta), that their eq. (4) has. That kind of
selection and observed dependency on the parameter theta has nothing to
do with the Bell inequality. The B.I. is a purely enumerative
constraint on the _entire_ data set (cf. [2] for one example of
enumerative formulation of Bell inequalities, which puts into its most
striking and pure form), of the same kind as the pigeonhole principle
by which if you have, say, 10 pigeons and 9 holes, there will be
necessarily one hole with two pigeons. Post-selecting a proper subset
of these 9 holes, say 5 holes, and pointing out that these 5 holes in
the subset all have single pigeon, doesn't in any way violate the
previous conclusion of the pigeonhole principle that in the 10+9 setup
there will be a hole with two pigeons. The existence of a subset of 5
holes, each with a single pigeon, is irrelevant as an experimental
demonstration of a "violation" of pigeonhole principle.

To get the full picture of what magic powers this "perfect detector"
would have to have, you need to read the Glauber's derivation of his
n-detector correlation functions Gn(x1,x2,...) in [3], in particular
his leap from the eq. (5.2b) to (5.5), where he starts with predicting
the dynamics of n detectors, then gives up the prediction and simply
defines his Gn() to retain only a small subset (approx. 1/e^n fraction)
of the dynamical terms "we are interested in" ([1], p. 85). A "perfect
detector", the G-detector, which has to show the raw counts which
correlate as Gn(), thus show QO non-classical effects, would have to
subtract not only the dark rates for itself (a single detector), but it
would have to know how to subtract 'accidental coincidences' and all
instances of m-triggers, for any m <> n. In other words, the G-detector
must be networked with all the other detectors and it needs an input to
the data-base of other experiments on the same setup (e.g. so it can
subtract 'accidentals' which are measured separately, with signal
turned off). That's the kind of "perfect detector" you need to design.
The non-local subtractions/dropping of the terms in Glauber's eq (5.5)
at the theoretical level is accomplished on he operational level by the
standard QO subtractions & filtering of the experimental data.

The classical theories, such as SED, don't include these (often
implicit) QO filtering conventions built into their predictions.
Therefore, the SED predictions refer operationally to different kind of
counts than the G-counts and G-correlations of QO. All the claimed QO
"violations" of the classical predictions are based on the gimmick of
extracting (non-locally!) the G-counts from the counts and comparing
them to classical model of the setup which doesn't include the model of
the QO subtractions (the Glauber's conventions, which are normally
implicit in the QO publications). In SED, the only subtractions applied
by default to the detectors are local subtractions of the ZPF
contributions (which, unlike the Glauber's subtractions, require no
knowledge of the remote results and results of the supplementary
experiments). Therefore the "predictions" of SED will appear different
from QO "predictions", simply because they refer to different kinds of
"counts". Until you normalize the predictions of the two theories to
refer to the same kind of "counts" (the distinction which your paper
doesn't even acknowledge to exist, in contrast to [4],[5]), you can't
make a meaningful claim having shown QO predictions which excludes SED.
When you include Bell's inequalities & sub-poissonian counts, that kind
of (straw man) "violations claims" are dime a dozen.

You should also note that in your paper & the comments given here, you
are melding together the original SED (e.g. your ref [1] cites T.
Marshall's 1963 SED paper) with its variant Stochastic Optics (SO) used
to describe the QO phenomena (cf. [4] and references there). The
original SED (which goes back to Plank & Nernst) is a much more
ambitious project, trying to show that the Schrodinger & Dirac matter
fields equations are deducible from a stochastic process (which
involves only classical charged particles interacting with classical EM
field, with ZPF initial & boundary conditions). This SED project can be
considered a failure. The much less ambitious project of SO, takes for
granted the matter fields equations and simply uses the ZPF for initial
& boundary conditions for EM fields. In order to replicate the QO
predictions (which it does for all PDC based sources), it also models
the QO subtraction procedures, as illustrated in [5]. While it can't
replicate the QED radiative corrections, the SO does replicate all the
phenomena of Quantum Optics to which it was applied (e.g. any order
coincidence experiments using PDC, cascade, coherent and chaotic
sources). Even in the well regarded QO textbook of Yariv [6], the last
chapter is dedicated to showing how various 'non-classical' QO results
obtained earlier via QED methods, can be replicated by inclusion of ZPF
in the classical EM theory. In the introduction to this chapter (p.
703), Yariv notes: "Somewhat to my surprise, I found that by asking the
student to accept just _one_ result from quantum mechanics [vacuum
fluctuations as classical ZPF], it is possible to treat all the
above-mentioned phenomena classically and obtain results that agree
with those of quantum optics."

1. Z.Y. Ou, L. Mandel
   "Violation of Bell's Inequality and Classical Probability in a
Two-Photon Correlation Experiment" Phys. Rev. Lett. 61(1) pp 50-53
(1988).
   http://prola.aps.org/abstract/PRL/v61/i1/p50_1

http://puhep1.princeton.edu/~mcdonald/examples/QM/ou_prl_61_50_88.pdf

2. Louis Sica "Bell's inequalities I: An explanation for their
experimental violation"
   quant-ph/0101087  http://cul.arxiv.org/abs/quant-ph/0101087
   http://xxx.arxiv.cornell.edu/find/quant-ph/1/au:+sica/0/1/0/all/0/1

3. R. J. Glauber "Optical coherence and photon statistics"
   in Quantum Optics and Electronics, ed. C. de Witt-Morett, A.
Blandin, and C. Cohen-Tannoudji
   (Gordon and Breach, New York, 1965), pp. 63-185.
   For discussion & objections see:
   http://www.physicsforums.com/showpost.php?p=529314&postcount=16
   http://www.physicsforums.com/showpost.php?p=535516&postcount=61
   http://www.physicsforums.com/showpost.php?p=538215&postcount=73

4. T. Marshall, E. Santos "The myth of the photon"
   http://cul.arxiv.org/abs/quant-ph/9711046

5. A. Casado, T. Marshall, R. Risco-Delgado, E. Santos
   "A Local Hidden Variables Model for Experiments involving Photon
Pairs Produced in Parametric Down Conversion"
   http://cul.arxiv.org/abs/quant-ph/0202097
   See also:
   E. Santos "How photon detectors remove vacuum fluctuations"
   http://cul.arxiv.org/abs/quant-ph/0207073

6. A. Yariv "Optical Electronics in Modern Communications"
   5th Ed, 1997 Oxford Univ. Press.