Clearance isn't a local problem, it's a global problem. Board houses require that your clearances be several times their minimum feature size, so that if they have poor resolution in any one spot, there will be no locations on your board that get shorted.
It's a statistical thing. If their process can resolve 1 mil with 1 sigma error, that's pretty bad for traces spaced so closely. But at 5 mil, it gets very good, parts-per-million good. But there are quite many traces and gaps on the board as well, so that there might be thousands of such trials; widening it to 7 or 8 mil ensures that even those will remain clear. In addition, you have quite good confidence that the voltage rating of those gaps will meet the minimum requirements for your design (functional insulation), or for safety purposes (UL/CE rated).
The most common failure I've seen, in board fabrication, is a very different kind of error: rather than a normally distributed error in the print, the most common seems to be relatively large (~mm) blemishes, presumably dust/contamination, or smears in the photographic process. There might be only a few such events per panel, and if one happens to span between traces, that board must be rejected. (If it happens to be over an open area, they might leave it, or scrape off the extra copper. If it happens over a filled area, no one's the wiser because it's already filled.)
Tim