The issues you give make sense when your datastructure grows very large or contains hash-unfriendly data, but it seems a plain hash table would still be appropr

The issues you give make sense when your datastructure grows very large or contains hash-unfriendly data, but it seems a plain hash table would still be appropriate in day-to-day scenarios. A simple int→int mapping is very fast in a hash table, and even string keyed hash tables are near unbeatable if you can reliably cache the hashes. The cases Radix trees win seem fairly small, as do the times you need incremental resizing.