I'm guessing here, but might that have to do with the way Python is executed on windows? If I remember correctly, Python programs are executed similarly to Java programs in that they rely on an interpreter to be present on the system. And if I remember correctly, most *nix OSes have an interpreter in their standard install, which windows doesn't, so Calibre has to bring it's own, which doesn't seem to be as fast. Does that make sense to anyone?
That being said, I've never tried importing hundreds of books, but I'd kind of expect the program to use whatever CPU time it could get its hands on to do tasks like importing or converting. On the other hand, taking hours to import books does seem to be a very long time, and 60-90% utilization when just running the UI is excessive, but, I believe, not a programming fault, but caused by the choice of Python as a programming language.
And lastly, for what it's worth, my Calibre installation (with a library of about 200 books) takes anywhere from 10-30% CPU when just jumping from one book to another. My system specs are roughly equal to yours.