Pythia version: 6.4
Pythia tune: tuneA
Filter: E_T sum for either half of the detector > 0, 2, and 4 GeV
Goal: 10M filtered events for each threshold.

I ran an initial batch of 1M filtered events with E_T > 2 GeV for quality check. Of these, I found 134 prompt photona in the FCS, not counting internal brems. 
I applied isolation cuts based on both the EMCal and HCal clusters (E > 0.5 GeV)with varying cone radius. 
At this point, the reconstruction is working on the entire detector without introducing artificial boundaries. 

Fig. 1. Ditribution of all photons vs. ones used as prompt photon candidates (E >5 GeV && fiducial cut around the edges)

Isolation cut is applied to all clusters found, with the minimum energy at 0.5 GeV, set by the reconstruction.

Fig. 2. All thrown prompt photons, no isolation cuts, and then r=0.3, 0.5, and 0.7.  

Fig. 3. Reconstruction efficiency and the effect of isolation cuts

With r=0.7, there are 22 photons with pT > 2.0 GeV, and 10 with pT > 2.5 GeV.

Prompt photon signal / everything ratio
r = 0.7, pT > 2.0 GeV : ~12%
r = 0.7, pT > 2.5 GeV : ~19%

Prompt photon signal retention fraction
r = 0.7, pT > 2.0 GeV : ~59%
r = 0.7, pT > 2.5 GeV : ~62%

Background rejection fraction
r = 0.7, pT > 2.0 GeV : ~88%
r = 0.7, pT > 2.5 GeV : ~87%