Membuster seems having several large CachedScript / CachedCSSStyleSheet Vectors, this sounds strange to me since it should be (hopefully) mapped as a mmap-ed file. My guess is something pathological happens, but need to check what is preventing from caching for these resources. 1. It is possible that they are non-cached due to the right policy. In this case, we cannot get improvement. 2. It is possible that Membuster's memory-capturing timing is super quick and we get this data before NetworkProcess's notification replaces the content with mmap-ed file, this is possible. But on the other hand, 1. It is possible that this is not cached due to wrong policy. We should double-check. 2. It is possible that this is not replaced with mmap-ed content due to wrong reason. We should double-check.
It turned out that a lot of resources in old Membuster is rejected in NetworkCache because it does not include any expiration headers at all. So freshness becomes 0, and it is simply rejected. But I think this is because old Membuster content is old... It would be true that new Membuster content has correct expiration header etc. and it caches things appropriately.