`vkEndCommandBuffer()` should be called from the thread that fills the command buffer

Gaming and Visualization Technologies APIs Vulkan
1

Nvidia suggest that

Functions such as vkAllocateCommandBuffers(), vkBeginCommandBuffer(), and vkEndCommandBuffer() should be called from the thread that fills the command buffer. These calls take measurable time on CPU and therefore should not be collected in a specific thread.

Does this suggestion just means to call vkAllocateCommandBuffers(), vkBeginCommandBuffer(), and vkEndCommandBuffer() in parallel, since Nvidia also suggest:

Parallelize command buffer recording, image and buffer creation, descriptor set updates, pipeline creation, and memory allocation / binding.

?

If my understanding is wrong, I didn’t get the point why vkAllocateCommandBuffers(), vkBeginCommandBuffer(), and vkEndCommandBuffer() should be called in the exact thread which does the recording (or does the word “fills” means recording commands?)