wow, thanks for the many replies!
I agree with, as many have pointed out, that my CPU is very old. I just can't get myself to buy a CPU+Motherboard+RAM combo - especially considering I bought extra RAM earlier this year, also thanks to you guys
Luckily, Chrispy_ mentioned getting a multi-core CPU for reasonably cheap might be the way to go! Or overclocking which I may or may not do... depends on how satisfied I am with a new GPU.
To help me decide I decided to take the TechReport approach and plot all the relevant graphics cards on the value plot. I did this by digitizing the data from three of this website's reviews. The dates of the reviews are given in the graph legend (I am to lazy to find the links again). Some of the data was omitted (due to cluttering) and I only showed up to $300. Since three different reviews obviously have different systems and games tested I had to scale each dataset relative to each other so that the FPS of one set for a particular card matches the FPS of the same card for another set. You will notice that there will be doubles of the same card plotted (but different color) which I kept there so one can see what data I used to link the different reviews. I also had to scale the price of all cards relative to the newest review (since cards get cheaper with time; finding the actual prices online would be best). The GTS 250 is the card that is the closest to the 8800GT so I included it by averaging the data from the September 2012 review and finding the cheapest unused price online. The arrows point to the cheapest card at a given FPS (the "obvious choice"). Finally, I kept the style of the graph the same as in this website so that it's easy to relate to.
GTX 650 Ti Boost seems like a significant enough improvement from the GTS 250 whereas the more expensive cards don't seem worthwhile to pursue for the little bit more in improvement. Do you think this graph is deceptive or more or less accurate? The only thing I see that is obviously off is that the 7850 and 7850 2GB aren't quite on top of each other which may be due to the uncertainty in the FPS or the fact that they were tested for different games or updated drivers. Another comment I have is that error bars (as determined by the standard deviation) would be an excellent addition to these review plots so that one can see which data is significantly different or not.
PS. sorry if the image is too large.Edit
: I updated the graph to reflect current prices on newegg.ca (if it wasn't found I used newegg.com, if not there then ebay.org). The dashed line is a logarithmic fit to the lowest price at a given fps cards. Top-left inset in the graph gives the current estimate of the value (FPS/$) of the card in the current market as determined from the derivative of the logarithmic fit. Green named cards are average valued across manufacturer (PNY, GIGABYTE, etc...) and from two different reviews scaled to fit the same FPS.