The basis for the +/- 3 sigma comes from normally distributed data. I strongly suggest to first evaluate your data as for distribution and statistical controls, making sure that the variation that you observe on the yield% is due to natural variation of the process or if its due to specific sources of variation that need to be investigated and corrected.

If you decide to arbitrarily change from 3sigma to 2 sigma, without evaluating distribution or statistical control, you could be wrongly identifying lots with low yield% when in fact, they were actually within the normal (expected) variation of the process.

It may be that some of those 150 batches had much lower yield% due to special sources of variation like validation batches, process deviations, etc., that could be contributing to the standard deviation of 0.37%.