Setting for a few lessons can be fine for quick wins where there are common areas of need and these are highly fluid and responsive to recent assessment. We still do this now and again for a lesson or two at a time.
But as a rolling practice, I think it's a big mistake in primary (and there's a lot of evidence against its efficacy).
First of all, it doesn't cut down the range of abilities in the classroom all that much. Most children fall into a bell curve which means that in the majority of cohorts the lower-middles of the lower set and the higher-middles of the high set would be closer to each other in general aptitudes than the lower-middles with the low-lows or the higher-middles with the high-highs. They might be stronger or weaker in different areas but with the average primary cohort size, you'd only have two or maximum three groups -- far too blunt an instrument to make much of an impact and an impediment to the majority of children around the middle.
I also found that moving between sets was very difficult because they would be taught different things making that jump harder. I've observed classes where they got around this by doing lessons in parallel but the lessons were so close it hardly seemed worth setting in the first place.
It's also not necessarily a recommendation that lower-attaining pupils are more comfortable being set -- yes, they may find the work less challenging but, to paraphrase Bart Simpsons, you can't catch up by going slower. I would guess most of these children actually need extra intervention and scaffolding to keep up with the bulk of the group.
Finally, in primary, the huge benefit is having one teacher know the children really well. In my experience, the benefits of knowing the children well generally outweigh the difficulties thrown up by a large ability range.
