And it is also the physical difference in how many pins the socket has. The 1366 is more for enthusiasts or more hard core users. The 1156 is more mainstream because of some lesser features, as stated by andy5174.
Lynnfield's IMC was complete garbage so yes a lot less.
Heck no. It's not that far behind a Bloomfield in bandwidth, and it's actually better in some latency measurements, depending heavily on the memory used of course. It's also a hair better than K10.5 in latency, and significantly better than K10.5 in bandwidth. I would be very surprised if you could find more than a tiny handful of programs that would be even close to memory limited on a Lynnfield.
I'll be honest, I don't really care about synthetics. Show me one application which performs significantly better on Bloomfield than Lynnfield because of the memory controller (and I don't mean a synthetic, I mean an actual, useful application). The benchmark you show there is more than sufficient for pretty much any modern program. It can easily keep all 4 cores fed with nearly any code. I know some people like benchmarks for the sake of benchmarks, but as far as I'm concerned, the purpose of a memory controller on a CPU is to prevent the cores from being starved of data, and from everything I've seen, the Lynnfield memory controller does so quite well.
In terms of bandwidth? No, it wasn't. In terms of programs? It absolutely was. Both Lynnfield and Bloomfield perform almost identically clock-for-clock on most programs. This shows that as far as programs are concerned, both of them have equally good memory controllers, since almost no programs require the sheer bandwidth capability of the Bloomfield memory controller.