bad scaling in `map`

```
julia> using Distributed

julia> addprocs(2);

julia> @everywhere using DistributedArrays

julia> a = fill(1000,10);

julia> da = distribute(a);

julia> @time map(x->rand(x,x)^2, a);
0.903241 seconds (63.61 k allocations: 155.698 MiB, 29.39% gc time)

julia> @time map(x->rand(x,x)^2, da);
0.967328 seconds (776.84 k allocations: 38.713 MiB)
```
(first time compilation omitted)

Even though this is embarrassingly parallel, the distributed version is consistently around the same time or slower. I tried this in julia 0.3 and the distributed time is around 0.5 seconds, close to the expected ~2x speedup.

Reproduce the reported timings using Distributed, addprocs(2), distribute, and map as shown in the issue. Trace the map and distribute entry points to find why the distributed call does not improve on the serial call. Done means the cause is identified and the distributed benchmark shows the expected scaling, with regression coverage if an existing test location is found.

Vom Indexierungsmodell aus dem Issue-Text verfasst.