The intrinsic dimensionality of a dataset is also relevant here.

The M-Tree is one of my favorite indexes. It works with data that's embedded in infinite dimensional spaces (sometimes; it's bumping up against an impossibility result that's sketched in a sibling comment).