I think you may want to reengineer what is happening a bit.

Basically what im thinking is that you have a manager script that prepares all the appropriate stuff, and then spawns a new perl to actually run the engineer code. Thus the engineer code runs in a seperate process under a seperate perl every time. Youll get some slowdown from marshalling data between the manager script and the worker script, but it should be workable and overall resolve the problem of unexpected use of $&.

Also with regard to your library usage, surely you can do a grep of your lib directories and simply remove or modify any library that has the offending vars in it. Obviously it would make your system non-standard, but it sounds like you have somewhat unusual requirements.