By caching the result of the first call to ThreadAlloc::get(), we were always hitting a code path that should be hit once per thread in normal operation.
By caching the result of the first call to ThreadAlloc::get(), we were always hitting a code path that should be hit once per thread in normal operation.