"I'd suggest you look at the existing Java or .Net Regexp objects and model your API on theirs."

Good point. I know Java supports RegExp now, but it is pretty weak comparing to Perl's. That's where my idea came in. However, it would be a good idea to make the API's close to Java's native ones.

It is even better to carefully compare Java's regexp and Perl's, and see in details what Perl can do to strengthen Java's. Maybe only making calls to the server when Java's native RegExp cannot handle the situation (this would in a way take care of itub's performance concern)...