JSON from SPARQL for multiple assertions

In the following:

_:a    ex:name   "Robert" .
_:a    ex:name   "Bobby" .

select ?name
where {?a ex:name ?name}

should the JSON output have a binding per name assertion or gather  
the name values in one binding? - one binding is much easier to  
process. If the values are gathered, should the literals be in one  
array or an array of values? In other words:

     "bindings": [
       {
         "name": { "type": "literal" , "value": "Robert" }
       } ,
       {
         "name": { "type": "literal" , "value": "Bobby" }
       } ,

		OR

     "bindings": [
       {
         "name": { "type": "literal" , "value": ["Bobby", "Robert"] }
       } ,
                    OR

     "bindings": [
       {
         "name": [{ "type": "literal" , "value": "Bobby" }, { "type":  
"literal" , "value": "Robert" }]
       } ,