Yes, you just have to spoof locations near the boundaries of the quanta. If they are less than 1km then test locations until they are in the 1km-3km group but you will know they are actually 1km away. Repeat between 3 and a dozen times at different angles and you can still determine a location.
Wouldn't eventually samples of arbitrary precision have their "range donuts" overlap heavily at the user's location?
This could probably be automated to track in real-time someone's position.