Yes it's definitely getting held up. The 595 rate I measured before is the maximum throughput I've been able to measure coming out of a splitter. the extra resources on the input side get held up as a result.
You need a really high framerate to see it stop though because the pause on the input belt is *brief*, or put another belt with a proper 780/minute flow next to it so you can compare the two belts moving.
The effect is a lot easier to see with mark 3 belts.
I've built a test rig to showcase an easy way to reproduce this issue. One large storage container, with two belts with the bottom one going through a splitter.
This is one large storage linking to two small storages stacked next to it.
After building everything, I placed exactly 2000 concrete in the large storage, then took this picture.
The bottom storage received 883 concrete. The top one received 1117.
Re-running the test without the splitter gave me the expected 1000/1000 split.