Btree statistics dominate the memory of an open dhandle, about 83KB, and disaggregated storage roughly doubles how many of those dhandles exist. The extra copies exist so concurrent sessions update different cache lines. Sys-perf on the disagg and ASC builds showed that fewer statistic buckets has little or no impact, and cutting the per-dhandle buckets would reclaim about 72KB on each open btree dhandle.

Disagg: https://performance-analyzer.server-tig.prod.corp.mongodb.com/perf-analyzer-viz/?comparison_id=6ab2975bdfdd2d66d2180427&selected_tab=data-table&determination_filter=Improved&determination_filter=Regressed&percent_filter=0%7C%7C100&z_filter=0%7C%7C10&filter_type=Default

ASC: https://performance-analyzer.server-tig.prod.corp.mongodb.com/perf-analyzer-viz?comparison_id=6ab2d9784f94e47740d56170&selected_tab=data-table&determination_filter=Improved&determination_filter=Regressed&percent_filter=0%7C%7C100&z_filter=0%7C%7C10&filter_type=Default