Of course I used compressed suffixes, but it didn't help much as compared with hash tables.

you could find a PATRICIA tree has even better memory footprint than whatever hash table you use

I don't think you can demonstrate that Patricia has a better memory footprint than a good hashtable. In fact best radix trees are much worse.

Of course, all data structures share the DoS problem.

You can attack radix trees in terms of memory usage, while hash tables are prone to serious performance attacks.

In all cases, the attack degrades performance to linear search.

Not true. I don't think you understand these algorithms and O(n) complexity very well.

The question how a hash table degrades when attacked highly depends on how exactly conflict resolution is implemented. Among other methods is, for example, expansion and rehashing, and you can't tell for sure if it degrades to linear search or not.

And radix never degrades to linear search - never.

Of all the "mainstream" data structures, balanced binary trees are probably the hardest to attack this way.

They are prone to a very specific attack when you make the tree to re-balance often.