Okay, I think I have the tie broken, finally:

There is no explicit ZDC trigger available consistently in the runs plotted in http://drupal.star.bnl.gov/STAR/blog/rcorliss/2010/apr/06/bht3coin-vs-dijets .  In order to have a roughly independent trigger against which to compare BHT3coin and Dijet, I created 'ZDCsum', which requires the sum of the east and west 'attentuatedSum's is >20.  

To review:

 

BHT3coin requires the BHT3 trigger to fire, and requires the sum of the ADCs in the HT trigger patches opposite the highest tower in phi to exceed 97.  The total opening angle is 10, +/-5 from the patch exactly opposite the highest.  It includes any eta.

Dijet requires the L2jet trigger to fire, with the Dijet bit in the L2Result set

ZDCsum requires the sum of east and west AttentuatedSums to be greater than 20.  (I admit I don't know what 'attentuated' means.)

The resulting ratios are plotted in figure 1.  The region near ~250 is relatively flat in the ratio nBHT3coin/nZDCsum (blue), but in the nDijet/nZDCsum (red) there is a bump of roughly a factor of two.  I conclude that the BHT3coin trigger is behaving normally in this region.  Additionally, we can see that other runs with outlying nBHT3coin/nDijet are caused some by dips in the blue, and some by dips in the red.  A diagnosis of each run follows.

FIgure 1, a comparison of various ratios of BHT3coin,ZDCsum, and Dijet, as computed from the largest mudst from each run.  It is assumed that nBHT3coin/nDijet should be flat under ideal conditions, and hence that any deviation from ~5 means either the BHT3coin trigger or the Dijet trigger has a problem that would invalidate its use.  The primary concern is the set of runs before ~250 where the ratio drops from ~5 to ~2.5.  By comparing the other two ratios, we can see that it is a shift in nDijet, and not nBHT3coin, that causes the shift.

Using a threshold of nBHT3coin/nDijet<3.0, there are 23 runs labeled bad from this set (labeled in figs 2 and 3, as well as the 'long version' attached at the bottom of the page).  If the discrepancy comes from a large shift in the nDijet/nZDC ratio, it is assumed to be recoverable, otherwise I recommend that the runs be removed from the runlist.

Figure 2, A closer look at the early section of figure 1, with any runs below the threshold of nBHT3coin/nDijet=3.0 labeled.  The vertical lines indicate a new fill.

Figure 3, A closer look at the later plateau region of figure 1, with runs below threshold labeled as before.

The overall structure seems to be both the ratios starting low at the start of the fill, then increasing over the course of the fill, then dropping again at the start of the next fill, always in tandem.  

R10087110, ratio=2.750000.  Dijet spikes high here.  Keep.

R10095120 through R10096027, ratio~2.5.  Unless something strange was done to the ZDC in these runs, Dijet spikes while BHT3 remains consistent with the previous fill.  Keep, but investigate.

R10096139, ratio=1.658915.  Dijet seems high, but BHT3 is also much lower than earlier fills.  Remove.

R10096140, ratio=2.003333.  Same as previous.  The Dijet spike seems to end here, so the remaining runs in the first half of this fill are likely to be okay.  

R10097038, R10097044, ratio~1.9.  Dijet spikes high here.  Keep.

R10097086, R10097088,R10097090, various ratios.  BHT3coin spikes low here, in a strange alternating pattern.  Remove.

The bottom line is that we have agreement of the Dijet and BHT3coin triggers over the bulk of the runs, and can reasonably identify the culprit in runs where they disagree.  I propose we remove runs 10096139, 10096140, 10097086, 10097088, and 10097090, which seem to show problems in the BHT3coin trigger.  Together, these correspond to 36.1 nb^-1, 0.2% of the total luminosity. (runs 10096139 and 10096140 are 12.08 and 16.11 respectively.  The others are all ~3)