This is a very difficult prediction to make, but there is one factor often left out of mining cost estimations: cost of hardware. When amortized over their expected lifespans, the costs of various pieces of equipment can become a meaningfully large factor in Bitcoin mining costs. In the days to come, it seems likely that FPGA or ASIC miners will become more common since they offer tremendous power efficiency gains over GPUs. The "downside" to FPGA and ASIC mining is that it is also extremely expensive to obtain the initial equipment, which means that the cost of equipment amortized over its expected lifespan is higher than with GPU mining.
We're now seeing the first commercially available FPGA mining rigs become available, costing about $440 for a 100 MH/s rig consuming 6.8 watts of electricity. Here at my home in Las Vegas, NV it would cost about 55 cents per month to run one of these FPGA boards, and even if the device survived 10 years of operation before its death, amortized hardware cost would be $3.67 per month, absolutely dwarfing energy costs. The ATI Radeon HD 5830 based rigs I currently run cost about $6.58 per 100 MH/s in electricity alone, but a 10-year amortization only adds $0.96 per month per 100 MH/s to that cost. This means the buy-in for GPU mining is low but so is the profit and sustainability.
This puts my cost per coin for FPGAs at $2.45 and for GPUs at $7.54, though there is certainly room for improvement in those numbers and your mileage WILL vary. The important bit, however, is that FPGA mining is far more profitable regardless of hardware cost and is likely to be the next big thing, which does a lot for Bitcoin's overall energy footprint since FPGA mining consumes about 8% of the electricity GPU mining currently consumes. If FPGA mining took over right now, the Bitcoin network could grow to 12 times its current hash rate without seeing a net increase in energy consumption.
In short, Bitcoin is in its infancy still and so is mining. Just as most technologies begin at low efficiency and high cost and grow towards high efficiency and low costs, so too is Bitcoin mining. You are correct in your assumption that electricity use will likely grow with Bitcoin adoption, but the missing factor in that equation is the increasing energy efficiency of mining hardware.