well i ended up running a map reduce script on a large data set, it took 2 hours, and process over 500k records, and all it did was run a tally of the number of records, so with that, i would assume that very large search objects can be used in map reduce, that they are not limited to the 4k search results