FS cache is, at least for me, included in "i/o resources". But this utilization will occur for any consumer reading any partition from any topic from anywhere o

FS cache is, at least for me, included in "i/o resources". But this utilization will occur for any consumer reading any partition from any topic from anywhere other than the tail segment which isn't currently being read (or even the tail segment of a partition not being produced); it's not specific to partition 0. And I don't believe you'd gain anything by turning on a mirroring cluster rather than increasing the number of brokers in the same cluster; in both cases you're solving it by spreading the i/o load out more.