There are slight but measurable performance gains to be had from refactoring how tokens are handled internally by the tokens API. Right now, tokens are indexed in a sparse array, and traversal is done (nearly) linearly based on the range index, with some intelligent skips where possible. PR to follow.