From Martin Cohen:

Well, we do somewhat better with IRAC than with 2MASS!  I believe
this is because the OSV region suffers substantial extinction
far beyond that for a normal line of sight.  This depresses the
observed JHK counts.  You can see the plausibility of this by
noting that the offsets between observed and predicted counts
in JHK are smaller with increasing wavelength.  

These are all the identical style of calculation.  Every SKY
plot is the result of 4 coadded independent runs with latitudes
of -1.05, -0.55, -0.05, +0.45 for GLON=284.191.  In this manner
we sample the highest counts close to the plane in one of these
runs and that better assesses the true counts.  This is the best
I can do with the material now in hand, at all wavelengths. 

The general accord can be taken to show you are doing nothing
terribly wrong with point source extraction and you can use the
divergence of observed counts from predicted to assess your
completeness limits in each IRAC band.  For example, somewhere
near 13.5 mag in IRAC1, or ~1.1 mJy; probably no poorer than that
based on SKY.