What SPARQL queries would you like to run against the LOD cloud?

Legacy questions from answers.semanticweb.com
#1

If querying multiple repositories with SPARQL were easy/efficient, what queries would you like to ask that span multiple (LOD or other) datasets?

#2 
SELECT ?person ?blackmail_reason
WHERE {
  ?person a foaf:Person ;
    foo:ashamedOf [ dc:description ?blackmail_reason ] ;
    foo:bankAccount [ foo:currentBalance ?balance ] .   
}
ORDER BY DESC(?balance)
#3

If I can search Google for the string foo, I want to search all literal objects in the dataset collection for it as well:

SELECT ?s ?p ?o 
WHERE {
  ?s ?p ?o .
  FILTER (regex(?o, "foo","i")) .   # with or without "i"
}

My other ideas are for evaluating the traction of various modeling constructs, which would be good to know when I consider whether to use them. (They're also very inefficient, but this is a brainstorming exercise, right?)

Find all predicates with foo in their name: 

SELECT DISTINCT ?p
WHERE {
  ?s ?p ?o .
  FILTER (regex(str(?p), "foo","i")) . 

}

Find out how many times a given property was used:

SELECT (COUNT(?o) AS ?barCount)
WHERE {
  ?s foo:bar ?o . 
}

Find out how many times a given class was used:

SELECT (COUNT(?s) AS ?instanceCount)
WHERE {
  ?s a foo:SomeClass .
}

Find out all the properties ever used with a given class: 

SELECT DISTINCT ?p
WHERE {
  ?s a skos:Concept .
  ?s ?p ?o . 
}

Find all the classes that ever use a given property: 

SELECT DISTINCT ?ClassName
WHERE {
  ?s a ?ClassName . 
  ?s foo:property ?o . 
}

I imagine there would also be a lot of queries to find connections between things, in which a given variable was in the object position of one triple pattern and the subject of another. SPARQL 1.1 property paths would make this even more interesting, letting you look for networks of connections. And, once you started finding patterns and connections, CONSTRUCT queries would let you create your own triples that take advantage of the connections you found.