... and just to add another sprig of julep:

C:\>perl -E "my $input='#abcd#ef#ghi#jklmnopqrstuv'; my @chunks = spli
+t(/([a-z]{3})/,$input); 
    use Data::Dumper; say Dumper @chunks;"
$VAR1 = '#';
$VAR2 = 'abc';
$VAR3 = 'd#ef#';
$VAR4 = 'ghi';
$VAR5 = '#';
$VAR6 = 'jkl';
$VAR7 = '';
$VAR8 = 'mno';
$VAR9 = '';
$VAR10 = 'pqr';
$VAR11 = '';
$VAR12 = 'stu';
$VAR13 = 'v';
[download]

Whereas without the capturing parens:

C:\>perl -E "my $input='#abcd#ef#ghi#jklmnopqrstuv'; my @chunks = spli
+t(/[a-z]{3}/,$input); use Data::Dumper; say Dumper @chunks;"
$VAR1 = '#';
$VAR2 = 'd#ef#';
$VAR3 = '#';
$VAR4 = '';
$VAR5 = '';
$VAR6 = '';
$VAR7 = 'v';
[download]

As suggested above, a simple regex might serve you better, and with fewer "gotcha's" ... whereas, while this might be a legitimate approach for certain problem cases, it is most certainly likely to be a HUGE! PITA for some successor_dev).

Afterthought (i.e., UPDATE) re OP's title: You're not trying "to split a string based on character(s) length" (as your title says, because

• Only when dealing with unicode chars does the length of a character vary :-)

Rather, you are trying to split a string based on a the length of a sequence of characters within the string (AKA: 'a substring length').