I'm not an expert, but I would think that the difference between the two would be negligible, considering that either way you're creating a scaled image. It would seem the only difference would be the time it would take to scale from thumbnail versus scale from bitmap, and either way I think you're talking about nanoseconds.
Honestly, it would seem that you would want to use #1 simply for memory use, rather than loading from full size image and then down sampling, you're loading it from the thumbnail and sizing up if necessary.