#3313 implements buffer reuse and found that it's enough to bring the number of allocated buffers back to 0.6.13 levels. However, it does not implement a young gen buffer as that's added complexity and doesn't seem to be necessary to fix the regression at the moment.

A young gen buffer could, especially for index data, provide an additional savings in required memory and buffer allocations. It doesn't seem to be necessary for collecting enough of a region's uploads to accurately estimate the required size since the implemented system does a good enough job of that already. Reducing the number of buffer allocations, and sharing buffers of small regions in particular, may further improve frame time stability during chunk loading on problematic Intel drivers.

This issue is a followup of #3300, which contains further elaboration of this idea.