One of the best 'sign-post' papers I ever read, which convinced me which way to go, was by Scarpa et al. (2006) (see reference below). There have been more conclusive ones published since, but this was the one I happened to read first.
To make dark matter fit general relativity to the oberved galaxy rotations you have to assume that it stays spread out in a halo around the galaxy, and therefore does not have structure on small scales. Scarpa et al. looked at globular clusters which are small areas within the Milky Way, where the stars are arranged slightly more densely than in surrounding areas. They found that the globular clusters behaved like little galaxies: whenever their internal accelerations dropped below a critical acceleration, a0, their dynamics became non-Newtonian. Their crucial point was that you can't use dark matter to explain the anomalous dynamics of tiny globular clusters since to fit it to galaxies you've already specified it must spread out: you can't have it both ways.
Scarpa et al. also pointed out that the external acceleration on the globular clusters due to the galaxy was larger than a0, but the anomalous behaviour still occured when the internal accelerations dropped below a0. I'm happy to say that this points away from MoND, and towards MiHsC which relies on the mutual accelerations of nearby matter.
I still have this glorious paper and I wrote on my copy: “Brilliant stuff! Tells me which way to go :)”. It seems to have been ignored by most of the astrophysical community, but it shows that the dark matter idea is unworkable.
Scarpa, R., G. Marconi and R. Gilmozzi, 2006. Globular clusters as a test for gravity in the weak acceleration regime. Arxiv: 0601581v1.
McCulloch, M.E., 2012. Testing quantised inertia (MiHsC) on galactic scales. A&SS, 342, 575. Arxiv: 1207.7007.