As per my suggestion to Etc, I'd personally say record in HD. Everything you may gain from super sampling down from 4K will be more than lost by the relatively higher compression (compared to HD) that onboard on the A7s uses.
If however you were pushing out uncompressed to an external recorder, 4K to HD wins by a mile
That's not true the H264 codec as well as any intra codecs are not optimised for 4K and perform worse than what they do for 1080p
The downscaling theory is based on the resolution available and the theory that downscaling a 4:2:0 color space becomes 4:4:4
The first is true only for cameras without full sensor readout and the second is just not true
Well. I did some test.
I recorded in a green screen set to test Chroma Key and I also recorded a low light shot.
All clips recorded with the Pana12-35mm. Iso400. 1/50s. I shot 4 different clips in each test:
-4K 25p 100Mbs
-HD 50p 200Mbs
-HD 50p 100Mbs
-HD 50p 50Mbs
Curiously I didn't notice much difference in the chroma key test. They all look pretty much the same. I definitely have to do further tests.
Let's get into the low light tests. Some stuff came up that really don't make much sense to me.
All clips recorded with the camera on a tripod, same duration (5 secs), same scenario, lighting, aperture...
4K size: 55,1MB
HD200 size: 28,6MB
HD100 size: 55,8MB
HD50 size: 33MB
When I saw the HD200 size in my computer explorer I had to double check all the clips in the camera because I couldn't believe it.
4K has double resolution than HD200, but half the frame rate, and half the bitrate. So in a rough calculation I would expect the HD200 clip to double the size of the 4K file. But, since compression is not linearly proportional to resolution and it has to do a lot with spacial and temporal redundancy ( bitrates are variable too), I'm really not surprised by this data.
What really surprised me is when comparing HD200 with HD100!!!!!!. Has no explanation to me. All the clips are in the same container and are supposed to be compressed by the same codec!!!!
Anyway, those are just number so far.
So I imported all of the clips into Adobe AE and pushed the signal to the extreme. I applied the same correction to all clips, obviously. I also downscaled the 4K clip to HD after correction.

Same cropped area of each clip:
HD 50Mbs

100Mbs

HD 200Mbs

4K

_________________
Depending on the frame you analyze of each clip you might see slightly quality difference between HD clips, but to my eyes the all look pretty much the same junk.
I expected some evident difference between them, but once again compression algorithms are beyond my understanding and I guess that probably I would notice some difference in terms of quality under some other circumstances (high lights or high detailed images).
BUT what I really consider as fact is that the 4K clip has a much better behavior than any HD file. For me it really confirms what I read about the convenience of recording in 4K in the GH4 even when the final output is HD.
Take a close noise in the door and in the AC switches. Here is no so evident, but in my work computer I can also see a great difference in the greens gradient of the chroma board. Is much less noisy in the 4K clip.