Many times I wish I had TraceDeoptimization available in a product build just to find it's a develop flag. In general, deoptimization is a very expensive operation so the overhead of an additional check is negligible.

Although there is one place where it could matter:

StackValueCollection* interpretedVFrame::stack_data(bool expressions) const {

  InterpreterOopMap oop_mask;
  // oopmap for current bci
  if ((TraceDeoptimization && Verbose) JVMCI_ONLY( || PrintDeoptimizationDetails)) {
    methodHandle m_h(Thread::current(), method());
    OopMapCache::compute_one_oop_map(m_h, bci(), &oop_mask);
  } else {
    method()->mask_for(bci(), &oop_mask);
  }

This method is used by various stack walking code in the VM and it might affect performance. Doing performance evaluation should give us an answer.