When I've presented early hands-on preview on DPR Nikon SLR Lens Talk forum on February 15th '06. ( http://forums.dpreview.com/forums/read.asp?forum=1030&message=17192261) some folks had doubts about my high ranking given to Sigma 17-70. This lens compared favourably to more expensive EX Sigma 18-50 F2.8, even exceling in some areas. Sample to sample variation was the possible cause of relatively inferior 18-50 performance compared to cheap 17-70. On the other hand, 17-70 could be above average sample, "spoiling" further those odd comparison results. As I'm no longer in possesion of that 18-50 sample, I couldn't check how good or bad it was. But these days I had a chance to try no less than five 17-70 samples (including the one I used in original comparison). In order to perform precize evaluation, I've used my "magic tool", the Imatest. Resolution was measured at center and at 75% from the center to the pic edge, along with chromatic aberration. Testing was performed using Nikon D200 and my standard slanted edge test target. Focal length was changed from 17 to 70 mm and aperture was held at max opening (changing from f2.8 to f4.5 as the f.l. changed).
As expected, the sample I've used for my original 17-70 vs 18-50 test (named "Lens A" in the chart presented by red lines) is the best among all five tested. Center resolution never drops below 50 lp/mm, getting even better when stopped down. The only fault of this copy is the lowest CA score, but you can't get free lunch here, right?! SO, if the 18-50 was slightly below average sample, it is clear that it could compare inferior to good copy of 17-70.
How samples differ on a larger count you can check in this interesting article: http://www.lensrentals.com/blog/2011/11/are-zooms-are-always-sharper-at-one-extreme-or-the-other