Take two neighboring regions on a brightfield IHC slide, one at chromogen concentration c and the other at 2c. On an 8-bit scale the lighter region reads 100 and the darker one reads 25. The naïve average, (100 + 25) / 2 = 62.5, inverts through Beer-Lambert to a concentration that is not 1.5c. It's higher than that, because the exponential weights the darker pixel more heavily.
Convert to OD first. The lighter region has OD = −log(100/255) ≈ 0.41 and the darker one OD = −log(25/255) ≈ 1.01. Their average, (0.41 + 1.01) / 2 = 0.71, is the OD of concentration 1.5c, up to the path-length and extinction-coefficient constants.
The effect isn't small. Averaging RGB intensities instead of OD can be off by 20% or more when concentration varies appreciably across a region, and quantitative IHC scoring routinely relies on per-region averages. The RGB-to-OD conversion engineLoading... is the standard preparation: run it before any region-averaging engine, and the measurements correspond to the biology.