We've already seen the "half way" point to DDR5 in LPDDR4X. What has it shown us? LPDDR4X has significantly higher latencies than DDR4, but also higher bandwidth numbers. The result has been a few percentage points in improvement in iGPUs on processors that have limited thermal headroom, and barely moving the needle in other benchmarks on similarly thermally constrained systems. It DOES use less power doing it, so there's that. What we've never seen is an H-class APU paired with LPDDR4X ram running at it's maximum spec to test what it can really do. However, we HAVE seen desktop APUs with heavily overclocked DDR4 sticks up around the same bandwidth, but with better CL numbers than it, and, when paired with Vega8, there's not a whole lot of difference.
It's going to take moving to the next generation of iGPUs for the extra bandwidth to be truly important.
As for desktop tasks, as we have seen with Threadripper, with it's 4 channels of DDR4, there are only a few benchmarks, ones that are HEAVILY memory bus constrained, where it's 4 channels of ram and the resulting bandwidth, made a notable difference in performance over a nearly equivalent AM4 CPU, especially any difference that couldn't be at least partially explained by the enlarged power/thermal envelope of the package. Big caches and "enough bandwidth for most use cases" is largely already here. The extra throughput helps in more and more specific situations.