To be honest, that makes no sense. /m only impacts the meaning of ^ and the only '^'s in the regex you show are inside a character class and so aren't impacted by /m.
If this change fixed the mod_perl client, then it does remind me of mod_perl bugs I've seen before. Usually these bugs are heisenbugs, however, coming and going depending on how long a given Apache process has been running, though I guess I have seen some persistent ones (and rarely ones that persisted until the master Apache process was restarted).
In any case, bug(s) I've seen cause rather weird regex behavior. For example, /\d\d/g will sometimes fail on a string full of pairs of digits (and you'll see a node temporarily tagged as being written "never"). Or I've had several complex regexes that mod_perl refused to parse even though they were correct and worked outside of mod_perl. Changing to a different delimiter was the most common solution.
Although I don't understand your explanation of the workings of this bug, it certainly looks like a Perl bug if dropping /m does what you describe (note that Perl bugs are more likely to be visible in mod_perl because the Perl interpretter runs for so long). It sounds like perhaps the /m on the one regex is, in mod_perl, making the /^END\r\n/ behave as if it had a /m on it.