As it turns out, it seems the definition of a word as captured by \\w is unclear, and the following code extracts the proper sequence :

gsub(x="LCZBDT22.105",
     pattern="([A-Z]*\\d*)(\\.)(\\d+)", replacement = "\\3")

It's as if a word can not have an uppercase within it (and it kind of makes sense), so the string "LCZB" would not be captured at all and is repassed to the output of gsub, according to what the help page says :

Blockquote Elements of character vectors x which are not substituted will be returned unchanged

the "T" and the "." are matched as group 1 and 2 and are therefore not kept.

Maybe this can help some other R-base coders.