I appreciate your response and I agree with you. However, you don't really address my point, which is that statistics skew can cause variations in query planning. Your scenario assumes a fixed data size with a homogeneous (or at least "known") distribution of keys. In your original comment, you noted that adding a row "in the wrong place" can cause seeks to go to scans.
As I thought through it, I realized you can get a good sense of statistics skew by getting the counts and distinct counts of each value for your keys. If you have a lot of variation, then you may be able to reasonably expect query plan variations.
I was looking at my original post you responded to. I was talking about catastrophic changes to speed based on data and machine considerations. They are not the only causes of problems but are often overlooked because people forget that this stuff runs on real hardware with definite limits beyond the query plan statistics.
As I thought through it, I realized you can get a good sense of statistics skew by getting the counts and distinct counts of each value for your keys. If you have a lot of variation, then you may be able to reasonably expect query plan variations.