That’s correct – if you simply use PSRAM as a workspace and allot it in a LIFO manner there will be no need to relocate your code hence no need for threaded code.
Are you sure you don’t want GC though? A good compacting GC can give you a lot of freedom when combined with dynamic typing and stack bounds checking. You could even have your dictionary exist as relocatable and retargetable threaded-code words existing in a GC’ed heap, allowing non-linear redefinition of words in your dictionary