An article on upgrading PCs made me chuckle. The result was spending dollars per percent of improvement in performance. They have no clue because they assumed M$ and that other OS was part of the solution. They are the problem for performance.
Assume you have two PCs, one fairly recent with 1gB or more RAM and a decent hard drive and the other years older and slower. Both have that other OS. Both take 60s to boot and even another 30s to allow you to use the desktop after logging in. Perhaps 10s to open OpenOffice.org.
Install Debian GNU/Linux on both. On the newer machine do a full desktop installation and open the ports for XDMCP and X. On the older system do a minimal installation plus X. Add the line “X -query ipaddress of newerPC” without the quotes and ipaddress is the network address on your LAN for the newer PC to /etc/rc.local.
The older machine will boot in about 30s and open a login screen to the new PC. Logging in will take 5s. Once you have opened and closed OpenOffice.org it will take as little as 2s to open it again. So, you get a 5X performance improvement and can use it on two computers for $0 and some time.
Which do you think is the better return on investment? Imagine if you can run 30 older PCs this way. The return on investment just gets larger. You can even use LTSP on the newer machine so the older machines boot over the network and your effort drops further for greater benefit. I have been doing this for years in schools. It works. With Debian, setting up LTSP is trivial. A few commands does the trick. If you own the LAN for a lab you can create your own DHCP server automatically and be in business.
Even if you only have one PC, typically installing GNU/Linux on it will give 2X performance improvement for $0. That’s infiinite performance gain for $0, a bargain.