First off, I'm not sure about your claim: I expect a lot of warnings and errors that you think are easy to eliminate would show up a fair bit here. Second, as far as Padre is concerned, I suspect that's sufficient anyway ;-)
The only technical issue I have with Narveson's interesting idea is in how difficult it would be to automatically discern between real perl code, pseudo-perl-code, and poetic-code. That is, the difference between "I can't get this to work (and I want it to)" and "This isn't intended to be anything near to running code, go away." Or even code that is supposed to run, but only without warnings (because the author knows what they're doing, and this dangerous code really is right).
Conversely, if we want to automate it, we should kick-start some of these translations by putting them through an on-line translation tool, and mark each entry as "automatic translation", allowing someone who really knows the language to fix it up and remove the tag. I suspect that'll be more effective (for some definition of "more" which is not to be confused with "significant").