Is there any downside to image averaging when rendering?

So this is a question for those who are familiar with some of the technicalities behind cycles…
Does anyone know wether there are any cases in which averaging out the result of multiple renders at different seeds reduces the accuracy of the result compared to a single high sample rate render? For example, does the intensity of caustics get reproduced accurately? And do certain settings like the noise pattern affect the accuracy of the image at low sample counts?

From what i understand, it isn’t as optimized to average out multiple renders, but it’s fairly marginal. you aren’t gaining anything by averaging multiple renders, but you don’t lose much, other than your time and effort to average and stack the renders.

So, no real downsides (other than it’s a bit of hassle) but there are no upsides either.

It shouldn’t affect accuracy, however, each render will reshuffle the sampling pattern. Ex, instead of 3000 samples from a single stratified sequence, you have 1000 samples each from 3 totally different stratified sequences. It’s more random and less evenly spread, and tends to look worse (which is the whole reason the sampler uses a stratified instead of purely random distribution in the first place).

tl:dr, it’s equally accurate, but often appears slightly noisier.