Setting parisizemax when running multiple Pari/GP instances on one system?

[hermann@stamm-wilbrandt.de]

I determined sum of four squares for Mersenne primes M_2..M_29.
I used ternary quadratic form method, M_29 took 23:03:02h to compute on 
AMD 7950X CPU.

Then I switched to using
https://pari.math.u-bordeaux.fr/Scripts/foursquares.gp

which took only 200s for M_29, and 315/861/14,198 seconds for 
M_30/M_31/M_32.
I did computations sequentially on 32GB RAM 7950X CPU with
hermann@7950x:~$ grep "^parisize" /etc/gprc
parisizemax = 31G
parisize = 31G
hermann@7950x:~$

https://github.com/Hermann-SW/Mersenne_primes_sum4squares#foursquaresgp


Next I switched to old Intel Xeon 6126 dual CPU sytem.
7950X has single threaded Passmark rating of 4300, while 6126 has 2139 
(half):
https://www.cpubenchmark.net/cpu_lookup.php?cpu=Intel+Xeon+Gold+6126+%40+2.60GHz&id=3219&cpuCount=2

But that system has 192GB RAM, and so I configured
hermann@i6126:~$ grep "^parisize" /etc/gprc
parisizemax = 188G
parisize = 188G
hermann@i6126:~$

But then I started three instances of GP to compute M_33, M_34 and M_35 
in parallel.
After more than an hour I realized that one instance got killed.

But the other two instances are running fine after nearly 7h:

top - 22:08:00 up  8:10,  1 user,  load average: 2.00, 2.00, 2.00
Tasks: 499 total,   3 running, 496 sleeping,   0 stopped,   0 zombie
%Cpu(s):  4.0 us,  0.1 sy,  0.0 ni, 95.8 id,  0.0 wa,  0.0 hi,  0.0 si,  
0.0 st
MiB Mem : 192075.6 total,  11274.2 free, 180249.0 used,    552.4 
buff/cache
MiB Swap:   1392.9 total,   1352.7 free,     40.2 used.  10692.8 avail 
Mem

PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ 
COMMAND
2243 hermann   20   0  175.1g  87.6g   3544 R 100.0  46.7 402:00.00 gp
2261 hermann   20   0  175.1g  87.6g   3736 R 100.0  46.7 401:28.61 gp
2890 hermann   20   0    9652   4284   3244 R   0.3   0.0   0:19.43 top
...



My question is, whether running k GP instance on single system should 
set parisizemax to ramsize/k or less?


Regards,

Hermann.