4-Way Handshake vs PMKID

[Mem5]

$ time hashcat -m 2500 test.hccapx --nonce-error-corrections=0 digit08
...
real 0m9,898s
user 0m7,541s
sys 0m0,933s

$ time hashcat -m 16800 test.16800 digit08
...
real 0m5,127s
user 0m2,792s
sys 0m0,821s

looking faster, for me!

The main time for computation is in the PBKDF2 (99.99999%, well not exactly but you get the point), not anything after that. Therefore anything after that PBKDF2 can be seen as almost not existent.

So why mode 16800 is faster - almost 2 times - than mode 2500 if the main painpoint is PBKDF2 for both ?