The load absolutely can change when applied with a different voltage. Recall that P = (V^2)/R. Suppose that the load is purely resistive and is 5Ω. At 120V, the load would be (120^2)/5 = 2,880W. At 208V, it would be (208^2)/5 = 8,653W. Having said that, unless we're talking about a universal voltage electronic lighting ballast, a power supply of some sort, or a special motor, very few devices can safely operate over such a wide voltage range. In the preceding example, you can see that, at 208V, the load dissipates three times as much heat as it does at 120V. Are you certain that your 208V, single-phase loads can operate at 120V?
In AC systems, voltages and currents are vector quantities - that is, having both a magnitude and a phase angle. As such, you cannot use basic artihmetic to compute voltages, currents, and other values of interest.
Let's define three voltages, Vab, Vbc, and Vca as follows:
Vab = 208∠30° V
Vbc = 208∠-90° V
Vca = 208∠150° V
Let's also assume that Vab serves your 1A load, Vbc serves your 2A load, and Vca serves your 3A load. For the sake of simplicity, we can also assume that your loads are purely resistive (Meaning that the current will have the same phase angle as the voltage). The resultant currents on each phase are as follows:
Ia = 1∠30° + 3∠150° = 2.65∠130° A
Ib = 1∠30° + 2∠-90° = 1.732∠-60° A
Ic = 2∠-90° + 3∠150° = 2.65∠-169° A
A common mistake that people make is to assume that, once you know the current on each phase, you can simply multiply each current by the phase-to-neutral voltage to obtain the load for each phase. Doing that, in this case, would yield (2.65 * 120) + (1.732 * 120) + (2.65 * 120) = 844 VA. Rather, as you indicated your original post, the load is (208 * 1) + (208 * 2) + (208 * 3) = 1,248 VA.
You should take a look at
[URL unfurl="true"]http://en.wikipedia.org/wiki/Euclidean_vector[/url] when you have some free time.