"I hadn't known that modifying the GST in a thread was bad. Is that documented? "

After further research im not sure where that idea came from, possibly mixing up in my head the 5.005 threads and the 5.8+ ithreads. In my head it was things that disappeared in or changed memory pointers in the GST in one thread could mess with other threads that used the same variable. But recent reading supports the idea that there are no perl shared memory spaces between the GSTs in different threads