Meh, sometimes when you’re optimizing things you realize new limits. This is how you figure out what works and what doesn’t, you try. Simplifying failures down to “it was too big” and acting like this should be a priori obvious is silly. A small amount of stuff is always going to exist a few standard deviations away from the mean. Sometimes it’ll move the mean, sometimes it will fail. Backpressure from being too much of something is pretty natural.