To be fair, Windows 7 made some great strides, and ended up being faster than Vista on a lot of older hardware. However, that does tend to be the exception. The general perception of performance (at least in the areas of MS that I had contact with when I was working there) is that it's important to hold the line in performance, and it's not generally valuable to invest in actually improving performance over time. MS spends a lot of resources measuring performance very accurately, but not nearly enough resources in making appropriate architecture and design choices, or in making big performance improvement pushes.

My experience and observation is that performance improvements of substantial magnitude are often available for the taking in almost all software projects, but it requires a concerted effort to realize those improvements. MS instead spends its efforts on merely making sure that the software doesn't get slower by too much.