Moer pipelines is only one of the issues, you have issues like pipeline length too. The Intel architecture has long pipelines, because they're so long and memory's so slow (in comparison) they use a prediction algorithm to work out what code will come next, and they execute it out of order. If the prediction is wrong they have to throw away a bunch of pre-executed code, flush the pipeline, and start from scratch on the next actual instruction. AMD has shorter pipelines so the issue isn't so important. This is all really deep down inside the processor, and not really relevant to end users.
Having more pipelines is good for highly parallel tasks, multithreaded software benefits from it, just like it benefits from multiple cores. Other issues are cache size, levels of cache before you hit memory (L1, L2, L3 are all in use), etc.
I don't know of any 200MHz processors around right now, even bus speeds are faster than that. I agree that highly parallel computers running the right software on the right problem will be quicker on that sort of machine than a PC. Places like Cray made (and may still make) vector supercomputers that process massive amounts of data in parallel, doing the same thing to each.
My processor theory is a little rusty, so there could be things that aren't quite right in there.
What's really important these days is that what you need to do runs as quickly as you need. On my 3500+ it's not as fast as i'd like, I know a dual core would be better for doing multiple things at once, I wonder if single threaded software will run as fast on an X2 processor as on a standard single core. So it's really the "in practice" stuff i'm wondering about.