I call perl bug. The following also demonstrates the odd behavior. Interestingly, the substr(lc($str),0) in the second iteration is limited to the length of the (correct) substr(lc($str),0) in the first iteration.

e.g. with an argument of 'a:bc', the 'bc' is cut down to 'b' (the length of 'a'). For 'ab:cde' or 'ab:cdefgh', the 'cde' and 'cdefgh' are cut down to 'cd' (the length of 'ab').

#!/usr/bin/perl -l
use strict;
use warnings;
use Encode qw/_utf8_on/;
for my $str (split /:/, shift||'a:bc') {
    _utf8_on($str);
    print "$str\t", substr(lc($str), 0);
#   use Devel::Peek; Dump substr(lc($str),0);
}
[download]

For someone familiar w/ perlguts (not me), uncomment the Devel::Peek line.

UPDATE: Expected output for input of x:yz is:

x x
yz yz
[download]

, but due to bugginess, it's:

x x
yz y
[download]

Also, the problem presents in v5.8.7 linux, but not in v5.8.0 solaris, if those are helpful data points.