Applications should not have to deal with throttling calls to `RLocker()`

[jonhoo]

Currently, each call to DRWMutex.RLocker() causes the fairly expensive CPUID instruction (at least on some architechtures) to be called. The benchmarking tool avoids this to some extent by caching the lock across multiple iterations on the lock, but applications should not have to implement this logic themselves. Ideally, the drwmutex should handle this itself, though I'm not sure how it might do so... Suggestions are welcome.