Last week I got a new development machine (box 4) upon which I will build Microsoft Office. It's an Athlon 64 with a retarded amount (2 GB) of memory. It seems to be pretty fast; it's an upgrade from my 3 GHz Pentium IV from last year (box 2). I've been working on setting it up off and on since Monday. It takes a while to get all of the stuff needed to build Office up and running. Luckily, there's not much I have to do; once the machine has been "seeded" with the Office build environment, I basically just keep running an automated script and rebooting until everything magically works. If I had to do all of the things that this script does by hand, it would have taken days of concentrated effort. With the magic script, it's just days and days of not having to pay much attention to the machine.
This is my fourth machine. There's my "mail machine," box 1, which is everyone's crappy machine that they didn't have the heart to throw away. I have a very low tolerance for slow computers, so it sat in the corner unplugged until someone came by looking for spare screws and I let him cannibalize it. I just answer my mail and visit websites with the pre-beta builds of Office 12 on my secondary development machine, box 3, which isn't actually used to build Office.
Frankly, I think that it would be well worth Microsoft's money to just throw away those old machines and let people do their non-development tasks on something with at least 2 GHz or so. Time is money. I imagine that they'd make up the value of the machines in increased productivity in less than a year. But maybe not. I'll stick with my current setup. Sure, you don't need 2 GHz to write an email, but it doesn't hurt.
I bet they'd get a full return on investment in a lot less than a year if they'd improve the wireless network in our building, but that's a different story. I'm sure they're doing whatever they think is most cost-effective overall...