rasta provides an elegant solution to your problem, if you need to use split. I tend to avoid fancy things like lookahead assertions if possible, if only to simplify maintenance. Also, since I was curious, I thought I'd benchmark our two solutions.

my $data = "00:01:00Something here bla bla blaTYPE00:02:00".
           "Something here bla bla blaANOTHERTYPE00:03:00S".
           "omething here bla bla blaEVENMORETYPES";

use Benchmark;

timethese (100000,

{ 
  withsplit => sub {
        my @lines = split /(?=\d\d:\d\d:\d\d)/, $data;
                   },

  nosplit   => sub {
        my @lines = $data =~ /(\d{2}:\d{2}:\d{2}[^\d]*)/g;
                   }
}

);
[download]

...gives me...

Benchmark: timing 100000 iterations of nosplit, withsplit...
   nosplit:  6 wallclock secs ( 7.09 usr +  0.00 sys =  7.09 CPU) @ 14
+104.37/s (n=100000)
 withsplit: 13 wallclock secs (13.85 usr +  0.00 sys = 13.85 CPU) @ 72
+20.22/s (n=100000)
[download]

I don't know if it's split or the lookahead that's slowing things down, but I thought you might be interested in my results anyway.

-Bird