The best way to test an IP core

Typically what I recommend is testing the IP core using bus functional models first so that you'll be testing the IP in isolation. Then once those check out then switch to a software based testing using the Nios II core to perform more exhaustive testing. Normally when I'm using Nios II to do testing I only simulate the system if the test fails which often occurs early when you have a new IP core. The step of doing the BFM testing first helps reduce the number of cycles you have to do while doing software driven testing.

I'm don't really believe there is a single "right" approach as long as you get the coverage that you require. If that coverage happens to only require software based testing then you could stick to it. Some users can get all their coverage using BFMs without the need to use software based testing. For the type of stuff I design I like using a software driven methodology since I can write C code faster than test programs for BFMs but that's just me so often preference also plays a role.