OK, Im finally back.
I was doing a rebuild of my work-pc (where I do my lite-c work),
and it went horribly wrong when I connected to the network...

Bunsen:
1> It must just be cacheing either a large chunk, or it is the
OS trying to 'find' the file. (deep path ATM)

2> Then do you know if the default.c diag panel shows this memory usage?
I couldnt see any change...

3> I have since discovered CreateFile CAN read non-buffered, but it gets ugly.
You need your own buffer/chunksize to match the HDD clustersize(simplified description)

>> If I open the file using CreateFile, I can move the 'readfrom' position manually using the 'SetFilePointer' API.
That way I could openfile, read a bit, skip a bit, read a bit, etc, close. NEVER going backwards.

BUT, since then, Ive decided I want to use a zip-library in my app to reduce the file-size.
Buffering becomes a necessity then, so Ive dropped the whole no-buffer idea completely.