Strategy used by rabbitmq to store messages in RAM/Disk

Question

I am using RabbitMQ version 3.7.17 and Classic Queues Version 1 (CDv1). Publishing message type is set to "persistent". According to RabbitMQ official documentation, in CQv1 (which is applicable here), RabbitMQ starts paging messages (in RAM) to Disk only when a certain paging watermark threshold is breached. (which by default is 50% of the total high watermark i.e total memory assigned to RabbitMQ broker). For eg.The below configuration starts paging at 50% of memory used, and blocks publishers at 40%.

vm_memory_high_watermark_paging_ratio = 0.50
vm_memory_high_watermark.relative = 0.4

However, in the RabbitMQ Management UI page, it is observed that even when the paging ratio threshold is well below the set limits (defaults), there are disk writes happening still parallelly (143/s) at almost same rate as publishing to memory (172/s). Screenshot management UI

The following is not clear here -- Why are disk writes happening when memory paging threshold is still not breached?

Follow-up questions on the same:

1. If, let's say, disk write is applicable for every message, then why RAM consumption is also there/high. Can we benefit by removing redundancy and setting lower watermark paging ratio, since disk write is anyway there?
2. When consumers connect to this queue, how is the memory managed? To elaborate, lets say there is a single consumer consuming from this queue, with a consumer prefetch count = 1000 set. Then will the rabbit broker maintain 1000 msgs readily available in RAM, even if the applied watermark paging threshold couldn't hold those 1000 msgs? How is RAM-Disk swapping at the broker is managed, taking the consumer connection scenario into account?

Studied a lot of related documents on RabbitMQ official docs , but couldn't find a concrete explaination to these issues. Please share your insights. Kindly keep yours answers relevant to only RabbitMQ v3.7.17 & below using only classic queues v1.