Why does NIOS write redundant stuff to SDRAM?

The printf code almost certainly uses malloced memory - which is why you are seeing in in SDRAM.

I would certainly try to get the stack in on-chip memory - and to use tightly coupled on-chip memory so you don't need to allocate resources to a data cache or large i-cache (you still need an i-cache for the boot code and JTAG debug - unless you are loading the code over PCIe).

If you do nee careful placement of code/data in memory blocks. I'd suggest writing your own linker script from scratch - start from the documentation rather than the default script!