On Solaris Sparc, running volano w/SafepointALot, one time I ran into the following (running jvmg1 build 53):

assertion in synchronizer.cpp, 370
  assert(v_false, "fast_exit() should have handled this already

The stack trace shows:
  ObjectSynchronizer::slow_exit(obj=0xf436ec58, lock=0xf436ec58, _the_thread=0x3429d0)
  Runtime1::monitorexit(thread=0x3429d0, obj=0xf436ec58, lock=0xf436ec58)

assertion:
  if (lock->displaced_header() == NULL)
   the enter was recursive via UseHeavyMonitors so we are done already

note: at the time of examining the core, lock BasicLock::_displaced_header=0xc0e82

- so another thread must have inflated the lock.