I have further investigated the problem and it seems that using
ib_util_malloc instead of malloc to allocate space for the return
value resolves the issue.

I am confused. Why is malloc working on most systems, but not on some?
Reading tutorials for writing UDFs, I came to believe that
ib_util_malloc is required only if you are using Delphi or Borland
C++, not Visual C++ (which I do).

Any official statement regarding this?