As John suggested, I applied loose cuts on acceptance of physics and spectrometer, and from the graph you can see that this noisy shower counter is actually outside this range. I also studied how many events will lose if we turn off shower#66 in the replay, and without any acceptance cut but only cut on Shower.vs.PreShower (R.sh.asum_c+R.ps_sum_c>1500), I have only about 2.0% electrons missing. However, if we apply acceptance cuts, I believe we won't lose any electron if (by software)turing off shower#66. So my conclusion is we don't really have to fix this counters but I still can fix it since it is not very hard.
Figure 1