It gets down to tradeoffs between bandwidth, cost, and expandability. Option 2 might be the best thing for you at this point as it's higher bandwidth than option #1 and a "balanced" configuration (as opposed to option #3). We've been recommending a balanced configuration (same DIMM type, same population per channel) whenever possible, as configs such as your #3 may cause the effective bandwidth to drop down to option #2 levels (as measured by STREAM) due to how the memory controller works.
With regards to your UDIMM question, using UDIMMs with ECC will likely be a lower cost option for 1GB and 2GB DIMMs. You still get memory ECC, and you still can run with DDR3 1333 when populating 1 DIMM per channel - populating all 6 DIMMs (2 DIMMs per channel) will drop the memory down to 1066 MHz. Of course, you need to check with Dell on what they support and the costs.
Hope this helps! Ed