mimalloc's RSS is much higher than other allocators in mimalloc-bench tests

The following are results from a run of mimalloc-bench. As you can see, the RSS for mimalloc is much higher than the other allocators. (sometimes up to 10x!) Is there any way to reduce this? This was running on an M1 Pro on macOS Monterey 12.4. The version of mimalloc was v1.7.6

$ ../../bench.sh -r=1 -n=1 je mi tbb sys allt
[...snip...]
#------------------------------------------------------------------
# test    alloc   time  rss    user  sys  page-faults page-reclaims

cfrac       je    04.08 2896 4.07 0.00 0 374
cfrac       mi    02.70 22592 2.68 0.01 0 1595
cfrac       tbb   04.64 2736 4.63 0.00 0 362
cfrac       sys   04.40 2272 4.39 0.00 0 319

espresso    je    03.93 4912 3.91 0.01 0 507
espresso    mi    03.30 23888 3.28 0.01 0 1676
espresso    tbb   04.09 3376 4.08 0.01 0 402
espresso    sys   04.01 3680 3.98 0.03 0 410

barnes      je    01.36 58384 1.35 0.00 0 3844
barnes      mi    01.36 77888 1.35 0.01 0 5051
barnes      tbb   01.36 58096 1.35 0.00 0 3822
barnes      sys   01.35 57584 1.35 0.00 0 3776

larsonN-sized je    4.493 109200 49.60 0.19 0 12317
larsonN-sized mi    3.078 154960 49.55 0.23 0 13123
larsonN-sized tbb   8.038 111936 49.54 0.12 0 8496
larsonN-sized sys   21.649 44912 47.21 2.50 0 8328

mstressN    je    00.58 154144 1.52 1.14 0 267326
mstressN    mi    00.32 495616 1.33 0.08 0 31697
mstressN    tbb   00.34 166000 1.56 0.05 0 14754
mstressN    sys   00.54 281312 2.11 0.40 0 19094

rptestN     je    .483 66192 0.82 0.62 0 6720
rptestN     mi    .488 126224 0.76 0.64 0 8102
rptestN     tbb   .689 91728 0.84 0.43 0 5962
rptestN     sys   .707 74352 0.98 0.57 0 5365

gs          je    00.73 36256 0.71 0.02 0 2978
gs          mi    00.71 55648 0.69 0.02 0 4181
gs          tbb   00.71 38592 0.69 0.01 0 3156
gs          sys   00.71 42384 0.69 0.01 0 3346

alloc-test1 je    03.20 17728 3.18 0.02 0 1718
alloc-test1 mi    03.01 31120 2.98 0.03 0 2129
alloc-test1 tbb   04.17 11744 4.14 0.03 0 925
alloc-test1 sys   09.52 13984 9.44 0.07 0 1051

alloc-testN je    03.67 22112 34.21 0.09 0 2005
alloc-testN mi    03.41 33536 31.92 0.07 0 2288
alloc-testN tbb   04.77 14512 45.22 0.09 0 1106
alloc-testN sys   41.57 12752 395.75 0.82 0 985

xmalloc-testN je    0.471 97728 47.25 0.49 0 6344
xmalloc-testN mi    0.307 117440 46.92 0.56 0 7540
xmalloc-testN tbb   0.909 80384 44.75 1.23 0 5244
xmalloc-testN sys   1.025 46992 43.93 3.85 0 3182

cache-scratch1 je    00.87 8752 0.86 0.00 0 1157
cache-scratch1 mi    00.87 21808 0.86 0.00 0 1547
cache-scratch1 tbb   00.86 1872 0.85 0.00 0 308
cache-scratch1 sys   00.86 1456 0.85 0.00 0 269

cache-scratchN je    00.12 8944 1.02 0.00 0 1179
cache-scratchN mi    00.12 21968 1.02 0.00 0 1566
cache-scratchN tbb   00.11 2048 1.02 0.00 0 330
cache-scratchN sys   00.11 1616 1.02 0.00 0 288

glibc-simple je    03.31 2432 3.29 0.01 0 355
glibc-simple mi    01.47 22112 1.46 0.00 0 1568
glibc-simple tbb   03.32 2288 3.31 0.01 0 340
glibc-simple sys   03.35 1920 3.25 0.10 0 300

glibc-thread je    1.981 10320 18.87 0.05 0 848
glibc-thread mi    .773 29040 19.08 0.04 0 2000
glibc-thread tbb   2.500 7984 19.10 0.03 0 692
glibc-thread sys   4.467 6656 19.12 0.04 0 595

Inizia in mimalloc-bench con bench.sh -r=1 -n=1 je mi tbb sys allt e riproduci i valori di RSS sulla configurazione M1/macOS indicata. Confronta le esecuzioni degli allocator e analizza come il benchmark registra RSS; il lavoro è concluso quando viene spiegata la discrepanza e viene identificato un metodo testato per ridurre l’RSS di mimalloc, se fattibile.