They are generally the same. The Xeon will allow for using ECC memory, which does error checking to make sure there aren't any flipped bits. Many users are happy with ignoring this possibility. This discussion gets messy. Intel says that flipped bits happen every two days. I will be discussing this with Intel sometime in the next couple weeks. You can view some of the research here, if you are interested:
There will be a significant cost difference between the T1650 and the T3600, but there is only a 4% difference in speed between the two chips (the E5-1620 is 4% faster). So you will have to justify if that 4% increase is worth the extra money.
Actually, the Core i7-3770 or Xeon e3-1270v2 in the T1650 will be slightly faster than the E5-1620 in the T3600 - even though the clock speed is lower. The "Ivy Bridge" CPUs (E3v2) are a bit more efficient per clock cycle than the Sandy Briidge E CPU (E5).
It really comes down to how robust a system you want. The T3600 can be configured with six or even eight core CPUs, relatively high-end graphics cards, LOTS of ECC memory and multiple drives. For some users, these will be important considerations. For example, industrial designers requiring high-end graphics should invest in the T3600 with a FirePro w5000 or W7000 graphics card.
For the majority of SolidWorks users - even those to do some FEA and rendering - I think the lean and mean T1650 is a better value.
I'd take the passmark results with a grain of salt. Ivy Bridge is only 4% faster than Sandy Bridge, and with .2GHz increase in speed, you will see ~6% boost. Doing the math, 6% - 4% = 2%. So I say the E5-1620 is 2% faster (previously I said 4, but it looks like I may have been off).
Try the passmark single-threaded benchmark: http://www.cpubenchmark.net/singleThread.html
Interestingly enough, it still shows the i7-3770 has faster.
In the end, we are talking about very small numbers. Real-world evaulations would be so close it would be impossible to tell. Save the money and go with the cheaper system.