Niagara is a different architecture with a different feature set and can't be compared directly. I'm talking about x86 vs x86.
The point of Niagara is that a single physical core is considerably less powerful than a single Nehalem core, while Niagara has 8-way SMT per core, rather than 2-way, like Nehalem does.
Apparently in certain multitasking scenarios this makes perfect sense. Because as I said: "having more threads in the air at the same time is more important than the absolute amount of processing performance available to each thread".
Which already pre-emptively countered your arguments of "Yeah, but remember than in Thuban, each thread doesn't share its execution resources, with Hyper Threading, two threads shares everything like cache and execution engine resources, and they tends to fight each other for resources, causing cache trashing."
There's considerably more fighting for resources going on in Niagara than in Nehalem, and still it's a fine CPU.
The key to Niagara is not that its virtual cores are terribly fast. The key is that they can put so many of these cores on a single die.
Intel's HT is far less extreme, but the idea is similar: You can get a CPU with 8 or 12 simultaneous threads, something that no other x86 CPU can do (okay, with the exception of Magny Cours, but at the cost of two large, powerhungry chips. Intel could fit far more threads in that same size and power envelope, using HT).
That's the extra value of HT: you get more simultaneous threads at the same cost.