Are there any strings in your logfile array containing CSS which shouldn't be altered? If not, you could just go with

for (@filenames) {
    s/CSS//;
}

# or written with a statement modifier:
# s/CSS// for @filenames;
[download]

You could run the above code over your array and check for false positives, and extend the regular expression if so.

Since your strings are tokens concatenated with a dot, of which the last is always ErrorLog, you could anchor your regular expression at the end with $ and remove CSS from the token preceeding the last two:

for (@filenames) {
    s{           # substitute 
      CSS        # the string CSS 
      (          # and a capture consisting in
        \.       # a literal dot,
        [^\.]+   # any number of non-dots,
        \.       # another literal dot,
        ErrorLog # and the constant ErrorLog
      )          # (end of capture)
      $          # anchored at the end of the string
    }
    {$1}x;       # against the capture above
}
[download]

Or written more concisely with a terse explanation:

for (@filenames) {
    s/CSS(\.[^\.]+\.ErrorLog)$/$1/; # remove CSS from end of antepenul
+timate token
}
[download]

Perhaps more effective and faster, if you have only a few strings with CSS: look for CSS, if found, split, remove and join:

for (@filenames) {
    if (/CSS/) {
        my @t = split /\./;
        $t[-3] =~ s/CSS$//;
        $_ = join '.', @t;
    }
}
[download]