glibc 2.26 puts a small allocation cache in each thread
engineeringclibcmemoryperformanceThe new malloc tcache avoids arena locking for eligible cached chunks. Its size classes, refill behavior and per-thread retention explain both the fast path and its memory cost.