I don't mind buffers in continuous-flow systems where everything that goes into a buffer will eventually be consumed. Some of it's necessary (and unavoidable) in simply processing materials. And in some kinds of production -- gears for belts comes to mind -- actual chest stockpiles are useful because demand can be very spikey and the assemblers can easily outrun the capacity of a belt to feed them. Power supply and uranium processing are other good ones.
But the characteristics of the ideal buffer system are:
1. No dead zone. Any material that goes into the buffer can be extracted without requiring additional material to be added.
2. The things consuming that material have priority over filling the buffer further (there's no point in filling a buffer if something that could use those materials is being starved).
3. The buffer is sized such that filling it doesn't take forever. Train stops with unlimited chests, for example, can consume extraordinary quantities of material if you're not careful.
Requiring the entire belt to be backed up before starting a nuclear reactor breaks principles 1 and 2, and potentially 3 depending on how your U-235 production is doing. It works okay if you start from a full belt to begin with, but it isn't really properly a buffer, because if your fuel production gets interrupted all that fuel on the belts is useless because the circuits will see you don't have a full belt and thus not enable the inserters.
(Note: I have encountered people whose 'solution' to any given production issue is 'eh, stick a chest in there somewhere', including the sciences. Baffles me. Science is the one thing you have absolutely no need to buffer whatsoever (excepting space, I suppose), because it's produced continuously and consumed at exactly known rates!)