But compiling something major are just the baby steps. What if you have to run something over a distributed cluster? Or if it relies on dozens of libraries that were written for *nixes, without consideration for the quirks of Apple? What if you want to run part of your stuff "facing the outside world"? What if your machine (and code) must run for weeks?
I've only ever administered scientific clusters in linux, but from what I've heared from people that work as admins in "we use Apple" groups (and scarce they are), it must be absolute hell.
One of the major strengths of the Mac is that they play fairly well with *nix systems. I don't know anyone who uses Macs for clusters - but Macs make really marvelous client machines for clusters.
For "outward facing" code, the Mac Pro really isn't a server, and using any workstation as a server is pretty fail. As far as uptime in weeks, my (old style) Mac Pro would be surprised to learn that it can't run for weeks at a time, given it has in the past, hammering simulations the whole time.
The point is that they're excellent client machines. You can rig up code on your Mac in R, Python, etc. and then hand it over to the cluster fairly effortlessly, at least in my experience compared to Windows.
I can only speak for myself, but at my most recent position, when we replaced my Dell with a Mac, my productivity shot up.
What you say makes sense. Now that I think of it, most of the pain that has been described to me can be attributed to the misguided notion of a professor that _everything_ must run on apple machines.
What was also notable is that most of the problems seemed to appear on the admin side of things (running the "mac servers"), while for the users, meaning the scientists programming and crunching on the machines, everything worked just fine. That of course added to the admin's frustration, as nobody could understand why they were complaining about the perfectly fine Apple-centric setup.
So Mac for the scientist and Linux for the server might quite likely produce happyness for everyone.
I've only ever administered scientific clusters in linux, but from what I've heared from people that work as admins in "we use Apple" groups (and scarce they are), it must be absolute hell.