Ok - I know what the problem is - I just don't know a proper solution:)

Just creating and initialising a Perl interpreter when the C thread starts solves all the problems, even though this C thread has no interaction with Perl at all . It seems when malloc and co are redefined, they use pointers that are associated with the current thread. In other words, when you use the Perl header files (XSUB.h and I suspect some of the others) there HAS to be an Perl interpreter initialised on all threads if you use malloc and it's friends.

Ideally, I don't want a Perl interpreter in each C thread:) There must be some way to build this module to use a global wide malloc?