I'm a TA (one of four) for the course Harper is talking about in this post. As they say on reddit, ask me anything. I do not guarantee I'll answer. I'll check back in in the morning.
However, a reasonable proportion of students seem interested, and a handful are overtly enthusiastic about the course. If I had to guess I'd say it won't be an order-of-magnitude improvement over 212, but it will be significant one, if only because it will reach more students earlier in their undergraduate careers. However, only time will tell.
As for 212's reputation, you and I must run in different circles; I know plenty of FP enthusiasts who got their first taste in 212. I suspect a strong selection bias is at work here.
In long:
Haskell has much better concrete syntax, better library and tool support, a more active community, and is subjectively nicer to program in.
Haskell is lazy, SML eager. Haskell catches a lot of flack about this from SML fans. I think the typical arguments against laziness (makes it hard to reason about space usage; just a special-case of eagerness) are not false but overblown (it's harder but not that hard once you get used to it; and yes, you can simulate laziness in an eager language using thunks etc, but that's a Turing tarpit argument). However, in the end I'm not sure laziness is worth it.
Haskell is pure and SML is not. I don't mind the impurity of SML (or any other language), but I think Haskell's purity has some nice side-benefits for a high-level FP language (eg. allows more compiler optimizations). I'm also grateful to it for being the main reason Haskell has support for monads, which I think are a powerful unifying idea well worth their weight even if you don't need them to encapsulate I/O & side-effects.
The big difference is, of course, their module systems. Everything Harper says about Haskell in this regard (https://existentialtype.wordpress.com/2011/04/16/modules-mat...) is true, but I don't think it's as big of an issue as he makes out. Typeclasses give you 90% of what you want, in a much more concise package. I've perhaps wanted ML-style modules & functors once or twice when writing Haskell, and I wish I had typeclasses all the time when writing SML. I could say more on this topic, but this response is already overlong.