I'm sorry to have to correct myself within seconds of hitting the "send button":
I don't know the MS-Windows APIs and never looked for a testcase. Indeed, it needs to involve callbacks to show the difference AFAIK.
This applies to CLISP only, because ffcall library will fix the stack pointer for direct calls. But in other Lisps which might inline te call to the function function, the stack pointer may get messed up even in direct calls.
Regards, Jörg Höhle.