LM-opencl poor speed on GT 1030 with some masks

As reported by @fanto666 in https://github.com/openwall/john/issues/4871#issuecomment-971916061 our default benchmark mask shows much better speed than an all-?a mask:

% ~/src/debian/john/bleeding-jumbo/run/john -format=LM-opencl -session:lm1 -mask='?a?a?l?u?d?d?s?s' lm1
Device 1: GeForce GT 1030
Using default input encoding: UTF-8
Using default target encoding: CP850
Loaded 2 password hashes with no different salts (LM-opencl [DES BS OpenCL])
LWS=192 GWS=65536
Press 'q' or Ctrl-C to abort, almost any other key for status
0g 0:00:00:12 DONE (2021-11-17 20:41) 0g/s 885069Kp/s 885069Kc/s 1770MC/s A?AJ46|..A?AJ46|
Session completed.

% ~/src/debian/john/bleeding-jumbo/run/john -format=LM-opencl -session:lm1 -mask='?a?a?a?a?a?a?a?a' lm1
Device 1: GeForce GT 1030
Using default input encoding: UTF-8
Using default target encoding: CP850
Loaded 2 password hashes with no different salts (LM-opencl [DES BS OpenCL])
LWS=96 GWS=524288
Press 'q' or Ctrl-C to abort, almost any other key for status
0g 0:00:00:49 0.11% (ETA: 09:11:53) 0g/s 165375Kp/s 165375Kc/s 330750KC/s AP3XMNA..AP3XMNA
0g 0:00:00:53 0.12% (ETA: 09:25:24) 0g/s 162450Kp/s 162450Kc/s 324900KC/s A:%XNA..A:%XNA
Session aborted

Start by reproducing the two provided john commands with LM-opencl on the GeForce GT 1030 and compare their mask-dependent speeds. The issue mentions no source files or tests; done means identifying and correcting the reported performance discrepancy, then verifying the benchmark behavior against these commands.