No drastic change is required. I don't think I like any of the options you have presented as much as two others. Consider:

code
pod
__DATA__
data
[download]

The only "problem" with this approach is that Perl will take the time to read the POD and parse it to the extent that is required to find an "=end" tag. If the pod is really huge or you feel the script is taking too long to start up, you could also consider:

code
__DATA__
data
__END_OF_DATA__
pod
[download]

where you have to pick an "end of data" delimiter that is unambiguous for your particular data and fix your <DATA>-reading code to recognize it.

        - tye (but my friends call me "Tye")