Noticed this while looking at Leyden profiles. C2 seems to spend considerable time doing in this code:

void PhaseIdealLoop::Dominators() {
  ...
  NTarjan *ntarjan = NEW_RESOURCE_ARRAY(NTarjan,C->unique()+1);
  // Initialize _control field for fast reference
  int i;
  for( i= C->unique()-1; i>=0; i-- )
    ntarjan[i]._control = nullptr;

The disassembly shows this loop is fairly hot. Replacing the initialization with memset, while touching more memory, is apparently faster. We seem to touch a lot (all?) these structs later on, so pulling them to cache with memset is likely "free".