I found a satisfying solution. Instead of using a return type of Array[Typle4[...]], I used a case class to create a simple structure:

case class Result (
    f_int    : Int,
    f_string : String,
    f_float  : Float,
    f_ts     : Timestamp
);
def foo2(p1 :Integer, p2 :Integer) : Array[Result] =
{ 
   val result = Array(
       Result(1, "1", 1.1f, new Timestamp(2014, 4, 17, 1, 0, 0, 0)),
       Result(2, "2", 2.2f, new Timestamp(2014, 4, 17, 2, 0, 0, 0)),
       Result(3, "3", 3.3f, new Timestamp(2014, 4, 17, 3, 0, 0, 0))
   );
   return result;
}
sqlContext.udf.register("foo2", foo2 _); 
sqlContext.sql("""
select
    a.foo_output.f_int      as f_int
  , a.foo_output.f_float    as f_float
  , a.foo_output.f_string   as f_string
  , a.foo_output.f_ts       as f_datetime
from (select explode(foo2(1, 7)) as foo_output) a
""").show()

This seemed to give the desired output and is the same as pyspark.

I'm still curious as to how to explicitly return a array of tuples. The fact that I got it to work in pyspark lends evidence to the existence of a way to accomplish the same thing in scala/spark.

Any thoughts?