Theoretically your argument would make sense, but practically your data distribution should be uniform enough for the statistics collector to work for terabytes and terabytes.

I've seen query plans get messed up with partial indexes, but if you have a regular index, you needn't worry.