No, it is not triggered by sed.
In my case it was triggered by make.
The makefile was using "which" and "sed".
Hmm... it's very sad we can't properly spot the exact situation when it happens but if we are more or less sure it happens in larger builds, what about of compilation of freemint kernel in debug mode and try it again? In this way we could log every malloc(size) calls and in the same time we could log response from freemint kernel. So if you Mark or Keith or anyone have setup suitable for some large build, it is in my opinion worth of trying, especially in aranym environment where you can redirect all debug (trace) output to console -- if the crash occurs, we'll have exact log of all (k)mallocs and (k)frees and probably, with some luck, we will be able to simulate it from simple C test case.