I'm not a routing expert - because, contrary to the above, some of us do cycle using a preferred mode of direct navigation between points. 'Compelely useless' it is not.
But 70% success using OSM sounds quite plausible to me. And there would be no logic to the failures, because it would be mostly down to the Wiki- nature of OSM, the completely variable nature of the contributions.
The most obvious type of problem would be roads that haven't been joined correctly in the map, so that routing cannot flow from one to the other. Sometimes 1-way sections of road are incorrectly tagged. Very often there is no consistency about the status of a road - primary, secondary, tertiary, other, bridleway, etc - a long load will often have been mapped by a combination of 2 or more people, who might each have tagged their section differently. (This shouldn't happen, but anything is possible if there's minimal quality control.) Depending on the routing mode you have set the GPS to (car, cycle etc), these inconsistencies will affect it in different ways.
This is not to say that OSM isn't a good map - it is, but it does have these characteristics. When a problem is found, why doesn't someone go in and fix it? Well in some ways I think refining the map is a far more difficult proposition than constructing it in the first place (which is now mostly done). It may involve treading on toes, or invading an area that someone regards as 'their patch', etc etc.