From a user experience point of view inputting information would be easiest, if the most used keys are in close proximity, and common keyboard combinations are next to each other.
While the DVORAK layout is efficient for typing with ten fingers that lay on the home key row and thus have fast access to the most commonly used letters, the same logic does not apply for swipe keyboard input, where you have one pointer. In fact, having all the most commonly used letters in a row might be counter productive in this case, because users would have to constantly slide left and right with constraint to one dimension, which actually makes longer movement paths.
Ideally, the most used keys would situate in the center area of the keyboard, allowing for little movement for the most common letters and detouring to the sides for the less frequently used letters.
The way swipe keyboards work, technically, is by comparing movement end points to possible words, under consideration of letters that were swiped over between corner points. In this light, the way QWERTY distributes the keys to left and right hand actually helps making words and the letters they constituate of more identifyable, because their movements are more characteristic. Clustering all the heavily used letters in the center area might thus backfire from a technical point of view, because it makes words less identifiable as less "zig-zags" movement happens.
With these two extremes, users' ease of input on the one hand and technical ease of word detection on the other hand, only thorough testing can reveal which would be the most efficient layout. My wager is, however, that it's those two extremes that dictate the framework. I doubt that DVORAK is actually a good fit with regards to both, usability and technical compability.