For simple cases, you could use this regexp:

while( $string =~ m/\bstart:(.+?)\bstop:/g ) {
    print "$1\n";
}
[download]

...and to get the last match using that regexp, apply it like this:

my $lastmatch = ( $string =~ m/\bstart:(.+?)\bstop:/g )[ -1 ];
[download]

<update>
...or for a pure regexp approach:

if( $string =~ m/\bstart:(?!.+?(?:\bstart:))(.+?)\bstop:/ ) {
    print "$1\n";
}
[download]

...which only further punctuates the point I make in the next paragraph.
</update>

But that's not going to be very robust. What if your needs mature to the point that you can no longer guarantee that 'stop:' doesn't occur embedded within the portion of the string you're capturing? For example, if 'stop:' is wrapped in quotes, should it be treated as a delimiter, or as text? For a solution that will stand up to these sorts of complex strings, forget about hand crafting a masterful regular expression. The hard work has already been done, refined, debugged, tested, and proven. To take advantage of the work that's already been done, have a look at Text::Balanced.