Has anyone ever actually benchmarked vm.page-cluster = 0? Makes no sense to me to suggest a cpu is so bottlenecked that disabling read-ahead would actually help. If anything it would mitigate the decompression time if it guessed correctly as the work would already be done if left at the default of 3. Normally cpu is not bound when using zram because its quite low cpu anyway.
Either a local SMTP server (less used) or an external service (more common). The SMTP is configurable but I believe most used option is ssl smtp over port 587.