Well; if anything, I’d say such results would indicate that such picture should be removed, as if people tag according only to that picture, the inferred meaning would be wrong in 40% of the cases, so it is obviously not a very good picture if this sample of users is any representation of reality (that is, the situation/tagging it associates has ambiguous meanings).
Another issue is that changing what the EXACTLY SAME PICTURE means today compared to what it meant yesterday is horrible UX, guaranteed to produce much more misery than it could ever hope to solve.
In addition to creating obvious huge amount of confusion, changing what the people though some tag meant would invalidate many of pre-existing “bad” and all future “intermediate” uses of that tag of people, making those two values in combination with surface=paving_stones unusable for all future (unless it was accompanied with mechanical edits removing all such values from the map first and preferably contacting the editors who used that in the past about the new meaning)
So, IMHO - either keep the status quo (thus at least not disturbing how it is is used and thus making all existing “bad” and all future “intermediate” problematic), or (better, but much more work) completely remove that picture by replacing it with something else (preferably something that creates less confusion, e.g. whose size is more likely to be correctly interpreted how big and deep the gaps are in reality, so there is [hopefully significantly] more than 60% of the users which agrees what it actually represents - probably best procedure would be taking several pictures and putting them to vote to find best candidate, and then checking in another vote whether it produces better results then current 60% match).
Note: One should also take into important consideration that the few dozen people who read this thread and vote are actually insignificant minority compared to number of people who mapped that tag (many of whom has looked at the wiki in the past, or used solution like StreetComplete which had etc.)
So while voting here could be very valuable input to creating new tags/values, for already widely used tags it is much less so - as 99% of the old people will continue using in the ways they’ve already learnt to do in the past. As in practice, only new mapper (who never used that tag before), and few dozen participants in this thread (insignificant) would actually use the new meaning. And all the users who would like to use the values, would have to use at history modification dates and guess is it more likely that users meant old meaning or new meaning?
I suspect that many of the bad voters are cyclists? I agree that this (rather unusual, I think) surface is particularly uncomfortable for cyclists, so maybe if this is the surface of a highway=cycleway, it should be tagged smoothness=bad?
I think that smoothness should be kept as being useful to the most sensitive vehicle that still can still use that surface, as it means the most difference there.
For example:
- if you drive SUV, you don’t really care very much whether it is
good or excellent smoothness of that surface=asphalt. But if you use roller blades, you do care. So the difference between good and excellent should be based primarily on rollerblade users, not SUV users (or other less sensitive vehicles).
- Same for other values, e.g.
intermediate vs bad is mostly interesting to about trekking-class bicycle users - it is not interesting for e.g. rollerblade users (as they can’t really use that surface anyway), and it is not really interesting to e.g. tractor users (as it doesn’t make a difference in their lives at all). While SUV is somewhat affected there, that effect is much less then the effect for trekking bicycle user; so in that specific example trekking bicycle user should be the one for whom the difference between that two smoothness values is primarily focused on (as they are about the most sensitive to that specific difference).
Or should a subtag bicycle:smoothness=* be introduced to handle such exceptions?
To play devil’s advocate; why not invent motorcar:smoothness=* instead and keep smoothness=* bicycle-related? Afterall, there are likely more bicycles in the world than cars 
. But, jesting aside, I don’t think that would be good idea in either case, because:
- firstly,
bicycle:* (or motorcar:*) as a prefix to *:smoothness wouldn’t help much. There is as much difference between downhill MTB bicycle and a road racing bicycle, as there is between sports car motor vehicle and an ATV motor vehicle. So It would make as much sense as conflating “sports_car” and “ATV” and “tank” into single category motor_vehicle:smoothness – i.e. not any sense at all, really)
- secondly, I don’t think trying to chop it down into dozens of subtags (and it would have to go that way if it made sense - e.g.
sports_car:smoothness, suv:smoothness, atv:smoothness, mtb:smoothness, city_bike:smoothness, road_bike:smoothness, tank:smoothness…) would do any good, really. It would make complex situation even more complex, harder to tag, harder to user, and overall less likely to be useful, and all that without really solving any issue.
We should not go down the road of more and more detail, because most mappers would not be able to follow and stop mapping it at all. I like it that 
Maybe it would be even better if the router app could also read your thoughts, sense your mood and take into account how well you slept last night, and recommend the perfect route 