You must be thinking of the badly outdated “falling” rotation curve of DDO 154. That wasn’t very persuasive at the time, and was shown to be incorrect by THINGS a decade ago. DDO 154 has M*<Mgas and is well for by MOND with essentially zero freedom.
This is a microcosm of everything wrong with science today. We don’t like MOND, so dismiss it the instant one scrap of dodgy data looks slightly out of place (eg DF2). But we like CDM, so spot it endless amounts of special pleading. The bullet cluster is a good example of that.
The bullet cluster is a problem for both theories. In LCDM, the problem is that the collision velocity is impossibly in excess of the escape speed. So - enter he special pleading: hydrodynamics makes it look faster than it is. That’s true, but it only buys us 500 km/s. Need 1500.
In the model that works, that extra 1000 comes from non-adiabatic compression during the collision. Ok, fine, that can happen - but it is again very improbable. The odds of the bullet cluster existing in LCDM are small to zero. But we accept it anyway because cognitive dissonance
I expected someone to bring this up, so have handy this figure for clusters from my 2002 review with Sanders. That was before the bullet cluster was discovered. 
Note that MOND suffers a discrepancy in pretty much every cluster. This is a more serious problem than any one cluster, like the bullet. So here we have to invoke special pleading for MOND. Maybe there are unseen baryons there. I find that lame, but it is a logical possibility.
If we believe Big Bang nucleosynthesis, there have to be some dark baryons out here somewhere. tritonstation.wordpress.com/2016/09/09/wha…
So we suffer a global missing baryon problem. MOND suffers one in clusters. LCDM suffers one on every scale smaller than clusters. How is that better? Is it better to need two kinds of dark matter when you’ve already accepted one?
tritonstation.wordpress.com/2016/07/30/mis…
tritonstation.wordpress.com/2016/07/30/mis…
So let look at the cluster data again. This is a real problem for MOND, but not one we have to plead about anymore than we do for LCDM. But at least MOND makes a prediction. LCDM does not. Anything in excess of pure baryons is dark matter - no matter how much. Anything goes.
So maybe MOND is wrong, though I think we’ve been too quick to leap to that desired conclusion. This blinds us to the deeper question I’ve been asking for over two decades. Why does MOND gets ANY prediction right? Just saying “dark matter” is not a satisfactory answer.
The fact that this crazy theory comes as close to the mark as it sow is surely telling us something. Yet the only answer I ever hear is “MOND is wrong!” so I refuse to think about any implication related to it. That attitude guarantees failure. Which is where the field is at.
Technical point: you say “the Bullet Cluster (no lensing around the hot ISM, which should be the dominant mass in MOND)” This is not correct. IF MOND is right, as written, there has to be more mass there than the ICM gas. This was known before the bullet cluster was discovered.
The only new information that the bullet cluster provided is that whatever the unseen mass is, it is collisionless. I’d be inclined to write MOND off for needing dark baryons, had it not happened before: it took a long tine for us to realize the ICM outweighs the galaxies.
Indeed, the problem is worse in that the required mass is not distributed like the gas, and has to be concentrated towards the center. But that’s also true in LCDM, where the baryon fraction only reaches the cosmic mean near the virial radius in the biggest clusters.
This is one reason why I think we’re up the proverbial creek without a paddle. If I sound annoyed, it is because I tire of my fellow professionals quoting falsehoods at me. Saying the lensing should follow the gas just shows that - like many - you haven’t thought this through.
