FWIW, the following may give a slight improvement in speed, simplicity or generality (I will not bother to put it into file-processing form):

>perl -wMstrict -le
"for (@ARGV) {
   s{ (?<= [[:upper:]]) ([[:upper:]]+) }{\L$1}xmsg;
   print;
   }
"
"XYZZY"  "RUN4LIFE"  "RUN42LIFE CO"
"GENERAL WIDGET CO., INTL"  "MI-GO BRAIN CYLINDERS, IPTY."
Xyzzy
Run4Life
Run42Life Co
General Widget Co., Intl
Mi-Go Brain Cylinders, Ipty.
[download]

Supposed advantages:

• The  ([[:upper:]]+) quantified capture grabs as much as possible of whatever is to be lower-cased (rather than capturing a character at a time);
• The use of the  \L interpolation modifier eliminates the need for evaluation in the regex (i.e., no  //e regex modifier);
• The use of  [[:upper:]] and  \L bring locale fully into play (I think).