The full layout is over 32% less "effort"** than qwerty; Dvorak comes in at only 30%, and Colemak barely beats it with 38% (and even that advantage rests on the debatable "difficulty" of A and E in qwpr; without penalizing home-row pinkies, I think qwpr would actually beat Colemak). Qwpr achieves that while moving only 11 keys, and only 2 of them change fingers; Colemak moves 17 (over 50% more, and most change fingers), and Dvorak about 34. The proof is in the results: on a cold-turkey conversion (to a previous, harder-to-learn draft), I could make it to close to half my previous speed (which means starting to rock a few combos) in under a day, better than I could do after a full week of Dvorak.
Qwpr's letter layout (the 30 central keys) differs from qwerty by 5 changes, each of which "pays for itself" by a reduction in effort of over 5%:
1: swap E/semicolon (a 7% reduction in "effort"**); and semicolon/P (1% and puts semicolon back on the finger that's used to it).
2: swap T/F (around 7%; same finger)
3: swap N/J (~7%; same finger)
4: swap I/K (6%; same finger. Combine with step 3 if you want to only relearn "ing" once)
5: swap L/O (~6%; same finger)
** All numeric measures of effort here are based on the carpalx system. The system includes some somewhat-subjective calls but it gives a good ballpark. In fact, I think that it's if anything undercounting the benefits of this layout; to me, because carpalx counts row and column effects as linearly seperable, it ends up counting home-row pinky strokes [A and E in qwpr] as too hard, and off-home-row pinky strokes [L in Dvorak] as too easy.