This behaviour is the same on two different OS/version combinations:

PC #1:
Version: 6.4.7.2 (x86)
Build ID: 639b8ac485750d5696d7590a72ef1b496725cfb5
CPU threads: 8; OS: Windows 10.0 Build 19043; UI render: GL; VCL: win; 
Locale: nb-NO (nb_NO); UI-Language: en-US
Calc: threaded

PC #2:
Version: 7.0.6.2
Build ID: 00(Build:2)
CPU threads: 4; OS: Linux 5.12; UI render: default; VCL: gtk3
Locale: nb-NO (en_US.UTF-8); UI: en-US
Calc: threaded

Steps to reproduce:
* Open one of the two files attached.
* Select B coloumn and do a Search and replace
 - Select "Regular expression" and Search In Formulas.
 - In search field: ([:digit:][:digit:]\:[:digit:][:digit:]\:[:digit:][:digit:])\.([:digit:])
 - In replace field: $1,$2
 - Hit Replace all

Expected behavior :
This search and replace is meant to replace the dot with a comma, according to the current locale settings.
The cells should now yeld hour-minute-seconds and 1/100 seconds.

Observed behavior :
The result does not comply with the regex, but instead it strip away the two first numbers - say it looks like the outcome is forced to comply according to the formatting of the cell instead of what the regex expression tell it to.

In cell B2, that contain "05:46:55.97"
 - expected cell value to become "05:46:55,97" (but appear as "46:55,97" according to the time formatting)
 - But cell content actually becomes "46:55,97"