After mruby 3.4

parser and code generator independent from mrb_state (picoruby?)
iv/hash entry cache
method inline caching improvements (cache method lookup results)
more peephole optimization (if possible)
built-in profiler (method call tracing, stack profiling, detailed memory analysis)
improved REPL (mirb) features (syntax highlighting)
configurable memory pools (per-object-type, memory-constrained devices)
suspend/resume VM state (serialize/deserialize for power cycling)
CMake build support (better IDE integration, standard C tooling)

Things to do (Things we need to consider)