I would suggest a script level calculated dimension would work fine in this case.
1. create a field while loading the data as
ceil(rowno()/25) as clusters
2. when u reload u get a new field called "clusters" with value "1" for first 25 values, "2" for the next 25 and so on till the last row in ur table...
3. use this cluster as the dimension. in the bar chart..
4. u may even bucket them with some business rules like ascending order of sales(which can be done with a order by in scripting)
Hope this helps.
Unfortunately, the bars below are calculated on the fly by taking value/average(other related values). The linkage is actually by a synthetic key that was formed using intervalmatch. Without destroying the intervalmatch-created bridge table, I can't flatten my data structure to the point where I can push the cluster number into the script (and thus, into the fact table).
I've basically given up on this... it doesn't look supported in Qlikview unless you can let the script do the heavy lifting.
I don't urgently need a solution; I'll leave this question marked unanswered in the hopes that a future QV release supports "arbitrary clustering".