It is a bit more complicated. I work in the analytical division of well-recognized company. Most of our vendors design instrumentation to work with Windows. There are rarely drivers for other OS choices. Most is also designed with an over-emphasis on graphical user interfaces, the bane of reproducible research.
I see way too much abuse of spreadsheets. According to Baggerly and Coombes, part of the problems in the Duke scandal were caused by off-by one index errors with Excel. Similar spreadheet blunders arose in the recent Reinhart-Rogoff problem.
I hate Excel. It is hard to do simple things efficiently. Try and do a scatterplot with multiple series. How many keystrokes will it take? Once you get your analysis done and your report written with Word, how difficult is it to fix if the client wants to add one more sample? Then consider the changes in VBA. We have 3rd party code that are locked and won't even open on current versions of Excel.
Over the last few years, I migrated all of my back-end data processing to R/Sweave/LaTeX. For some projects I use markdown instead of LaTeX. Everything is scriptable, plays well with version control (code is mainly text files), and runs on Linux, Mac, and Windows. I use (and contribute) to Open Source whenever feasible. Solving problems is easier and I find community support better than most vendor support.
If I could get my hardware to play nice with Linux, I'd switch in a heartbeat. There is only one application I would miss - the debugger in Visual Studio. RStudio is pretty good at what it was designed for, but that does not include debugging the C++ code that needs to be written to speed up some computationally intensive parts...