I think the issue here, Jani, is that the closer the test target the greater will be the significance of any degree of misfocussing. Focussing at infinity using autofocus is probably going to allow greater accuracy with less hassle.
On the other hand, distant objects are not necessarily good targets for comparing resolution, due to often lower contrast.
Another issue which might complicate matters is the fact, apparently, that all lenses do not necessarily exhibit equal performance at all focussed distances. Lens 'A' might be betterthan lens 'B' at infinity whereas lens 'B' might be better than lens 'A' at a close distance of 2-3 metres, at the same f stop.
Having tested a few of my own lenses, I'm acutely aware that the slightest degree of misfocussing can ruin the results, particularly at wide apertures. For this reason I prefer to use test charts with progressively narrower lines and manually focus if necessary. On some part of the chart, and at an appropriate distance from the chart, the lines will be spaced so that they will cause aliasing and color aberrations which are clearly visible through the viewfinder.
When these artifact are at their most obvious, this represents the most accurate focussing I know of. Turn the focussing ring by the slightest degree and you can see such colorful aberrations diminish and/or disappear entirely.
It's also interesting that sometimes when using autofocus on such a chart (Norman Koren's is ideal) the autofocus will appear to be 'spot on' the first time the shutter is half depressed, slightly off-focus the second time and back on focus the third time.
